Blast furnace multi-element molten iron quality control method based on sub-modal division constraint optimization

Abstract

The invention relates to a blast furnace multi-element molten iron quality control method based on sub-modal division constraint optimization, which comprises the following steps: acquiring operation data in a blast furnace production process, constructing an original data set, cleaning, selecting an input variable and an output variable of a model by adopting a correlation analysis method, and calculating the quality of the blast furnace multi-element molten iron. Determining an input and output variable combination of the nonlinear dynamic characteristics in the blast furnace smelting process based on a time sequence relationship between the input and output variable data; based on the input and output variable combination, establishing a soft measurement prediction model by applying a back propagation neural network, and performing online prediction on the blast furnace molten iron quality index; and based on the soft measurement prediction model, the blast furnace molten iron quality is controlled by combining a feedback correction mechanism and a rolling optimization strategy, in the rolling optimization strategy process, a K-means clustering algorithm is introduced based on historical operation data to divide sub-modes, a constraint optimization space is determined through the current working condition, and the rolling optimization strategy is optimized. Therefore, the optimization process is always carried out in a safe range.

Description

Technical Field

[0001] The present invention relates to the technical field of blast furnace smelting process monitoring and control, and in particular to a blast furnace multivariate molten iron quality control method based on submodal partitioning constraint optimization. Background Art

[0002] During the blast furnace smelting process, the control of molten iron quality is crucial to the quality and production efficiency of steel products. Existing blast furnace control systems mainly rely on the monitoring and regulation of blast furnace molten iron quality-related variables and molten iron quality indicators. However, these systems often lack comprehensive consideration of the highly coupled relationship between molten iron quality control and equipment safe operation. In the actual production process, the blast furnace operating conditions are complex and changeable, and the control of molten iron quality indicators is not only affected by the current operating variables, but also closely related to the historical operating status of the blast furnace.

[0003] Existing technologies usually predict molten iron quality indicators online by building soft sensor prediction models, and implement control strategies based on the prediction results of soft sensor prediction models. However, these soft sensor prediction models often fail to fully capture the nonlinear dynamic characteristics of the blast furnace smelting process, and the control optimization process often ignores the historical data of blast furnace operation and the constraints of current operating conditions, resulting in the need to improve control accuracy and safety.

[0004] Therefore, the existing technology has deficiencies in the quality control of molten iron in blast furnaces, mainly reflected in the failure to comprehensively consider the highly coupled relationship between molten iron quality control and equipment safe operation, and the lack of comprehensive analysis of blast furnace historical data and current operating conditions. This limits the control accuracy and safety of the blast furnace smelting process, and affects the quality and production efficiency of steel products. In order to solve these problems, it is necessary to develop a blast furnace multivariate molten iron quality control method that can comprehensively consider multiple factors. Summary of the invention

[0005] The purpose of an embodiment of the present application is to provide a blast furnace multi-element molten iron quality control method based on sub-modal partitioning constraint optimization. This method enhances the control accuracy and safety of the blast furnace by constraining the optimization space to cope with the challenges brought about by dynamic changes in the blast furnace ironmaking process.

[0006] According to a first aspect of an embodiment of the present application, a blast furnace multivariate molten iron quality control method based on submodal partitioning constraint optimization is provided, comprising: S1: Acquire the operation data of the blast furnace production process to construct the original data set, including blast furnace molten iron quality related variables and molten iron quality indicators; S2: cleaning the original data set, and then using the correlation analysis method to select the input variables and output variables of the model, and determining the input and output variable combination of the nonlinear dynamic characteristics in the blast furnace smelting process based on the time series relationship between the input and output variable data; S3: Based on the input and output variable combination, a back propagation neural network is applied to establish a soft measurement prediction model for online prediction of blast furnace molten iron quality indicators; S4: Based on the soft measurement prediction model, the quality of blast furnace molten iron is controlled in combination with the feedback correction mechanism and the rolling optimization strategy. In the rolling optimization strategy process, the K-means clustering algorithm is introduced based on the historical operation data to divide the sub-modes, and the constrained optimization space is determined by the current operating conditions to ensure that the optimization process is always carried out within a safe range.

[0007] According to a second aspect of an embodiment of the present application, a blast furnace multivariate molten iron quality control device based on submodal partitioning constraint optimization is provided, comprising: The data acquisition module is used to obtain the operating data in the blast furnace production process to construct the original data set, including blast furnace molten iron quality related variables and molten iron quality indicators; A data processing module is used to clean the original data set, and then use a correlation analysis method to select the input variables and output variables of the model, and determine the input and output variable combination of the nonlinear dynamic characteristics in the blast furnace smelting process based on the time series relationship between the input and output variable data; A model building module, for establishing a soft measurement prediction model based on the input and output variable combination and applying a back propagation neural network, for online prediction of blast furnace molten iron quality indicators; A control module is used to control the quality of molten iron in the blast furnace based on the soft sensor prediction model, combined with a feedback correction mechanism and a rolling optimization strategy. In the rolling optimization strategy process, a K-means clustering algorithm is introduced based on historical operating data to divide the sub-modes, and the constrained optimization space is determined by the current operating conditions, thereby ensuring that the optimization process is always carried out within a safe range.

[0008] According to a third aspect of an embodiment of the present application, there is provided an electronic device, including: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

[0009] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0010] The technical solution provided by the embodiments of the present application may have the following beneficial effects: It can be seen from the above embodiments that the multivariate molten iron quality control method for blast furnaces based on submodal partition constraint optimization proposed in this application has a significant effect on improving the operating efficiency of blast furnaces and the control level of molten iron quality indicators. The present invention first provides a solid foundation for subsequent analysis by comprehensively collecting blast furnace molten iron quality related variables and molten iron quality indicator data in the blast furnace production process. Then, the variables are carefully screened using the data correlation analysis method to determine the core input parameters affecting the molten iron quality, and to determine the input variables and output variables of the model. This process effectively reduces the complexity of the model and improves the subsequent prediction efficiency. When constructing a soft measurement prediction model, the present invention combines time series analysis, and based on the screened input variables and output variables, determines the input and output variable combination that can accurately reflect the dynamic evolution in the blast furnace smelting process. Subsequently, a back propagation neural network is applied to construct a soft measurement prediction model, which realizes the online accurate prediction of blast furnace molten iron quality indicators, and provides timely and accurate data support for blast furnace operation adjustment. Afterwards, based on the soft measurement prediction model, the blast furnace molten iron quality is controlled in combination with the feedback correction mechanism and the rolling optimization strategy. The present invention introduces an innovative sub-modal division method into the rolling optimization strategy. The historical operation data of the blast furnace is used to divide the blast furnace operation status in detail through the K-means clustering algorithm, and different operating condition subsets are identified. According to the sub-modal of the current blast furnace operation status, the corresponding constrained optimization space is determined. This measure effectively limits the optimization process to a safe range and greatly reduces the safety risk in the control process.

[0011] In summary, the blast furnace multivariate molten iron quality control method based on submodal partitioning constraint optimization provided by the present invention not only realizes dynamic and accurate control of molten iron quality indicators blast furnace molten iron temperature and molten iron silicon content, but also significantly improves the operation efficiency of the blast furnace and the control level of molten iron quality indicators. This method provides important technical support and guarantee for optimizing blast furnace operation, and is of great significance for promoting the further development of blast furnace ironmaking technology.

[0012] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0014] Figure 1 It is a flow chart of a blast furnace multivariate molten iron quality control method based on submodal partitioning constraint optimization according to an exemplary embodiment.

[0015] Figure 2It is a schematic diagram of a blast furnace multivariate molten iron quality control method based on submodal partitioning constraint optimization according to an exemplary embodiment.

[0016] Figure 3 The figure is a diagram showing the modeling effect and prediction effect of a soft sensor prediction model based on blast furnace data according to an exemplary embodiment.

[0017] Figure 4 The figure is a control block diagram based on a soft sensor prediction model according to an exemplary embodiment.

[0018] Figure 5 It is a result diagram showing the effect of using sub-modal partitioning and clustering on blast furnace data according to an exemplary embodiment.

[0019] Figure 6 is a schematic diagram of a control input variable constraint space after being divided using sub-modes according to an exemplary embodiment.

[0020] Figure 7 It is a simulation effect diagram showing the setting value tracking effect and the control variable output using the constraint-added optimization predictive control according to an exemplary embodiment.

[0021] Figure 8 The diagram shows how to change sub-modal labels in a process of optimizing predictive control using constraints according to an exemplary embodiment.

[0022] Fig. 9 It is a simulation effect diagram showing the setting value tracking effect and the control variable output using unconstrained optimization predictive control according to an exemplary embodiment.

[0023] Fig.10 It is a block diagram of a blast furnace multi-element molten iron quality control device based on sub-modal partitioning constraint optimization according to an exemplary embodiment. DETAILED DESCRIPTION

[0024] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0025] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0026] Figure 1 The flowchart of a blast furnace multi-element molten iron quality control method based on sub-modal partitioning constraint optimization is shown according to an exemplary embodiment. The blast furnace multi-element molten iron quality control method based on sub-modal partitioning constraint optimization provided by an embodiment of the present invention may include the following steps: S1: Obtain the operating data during the blast furnace production process to construct the original data set, which includes blast furnace molten iron quality related variables and molten iron quality indicators, including the following sub-steps: S11: Acquire relevant production parameter process data from the historical operation database of the blast furnace; Specifically, we read relevant production parameter process data of a large blast furnace (2650m³ blast furnace) in China over a period of time from its database. These data cover 29 blast furnace molten iron quality related variables, including furnace top pressure, permeability index, furnace top temperature, and other important molten iron quality indicators, such as molten iron temperature and molten iron silicon content. By integrating these data, we constructed a comprehensive and detailed blast furnace dataset.

[0027] S12: Based on the relevant production parameter process data, the data are divided into blast furnace molten iron quality related variables and molten iron quality indicators according to the physical position of the variables in the blast furnace production process and their functional roles.

[0028] Specifically, variables related to blast furnace molten iron quality include process parameters such as oxygen enrichment flow, cold air flow, and hot air pressure, and molten iron quality indicators include molten iron temperature, molten iron silicon content, etc. Through this division method, the original data set is constructed to provide basic data support for subsequent modeling and analysis.

[0029] S2: Cleaning the original data set, and then using the correlation analysis method to select the input variables and output variables of the model, and determining the input and output variable combination of the nonlinear dynamic characteristics in the blast furnace smelting process based on the time series relationship between the input and output variable data, including the following sub-steps: S21: performing data cleaning on the original data set; Specifically, a data preprocessing process is performed on the original data set to ensure the accuracy and reliability of the data, including but not limited to the following methods: A. In order to solve the problem of missing values ​​in blast furnace data, the moving average method was used to fill in the missing values. This method effectively ensures the continuity and integrity of the data and provides solid data support for subsequent analysis.

[0030] B. In order to detect and remove outliers in the blast furnace data, the box plot method was used. Through strict detection steps, outliers were successfully identified and removed, thus ensuring the accuracy and validity of the data.

[0031] C. During the data processing, special attention was paid to the data synchronization problem at different sampling times. By collecting the variable data related to molten iron quality and the molten iron quality index data of the blast furnace, and importing them into professional data processing software, the linear interpolation method was used to convert these data into a time series with equal sampling intervals. This step ensures the continuity and consistency of the data, providing strong support for subsequent time series analysis and model building.

[0032] D. In order to eliminate the influence of different dimensions on data analysis, the maximum and minimum method was used for normalization. By converting all data into values ​​between 0 and 1, the accuracy and reliability of subsequent modeling and analysis were effectively improved.

[0033] S22: Based on the cleaned data set, the correlation coefficient between the blast furnace molten iron quality related variables and the molten iron quality indicators in the data set is maximized by using correlation analysis methods (including but not limited to typical correlation analysis using mathematical optimization methods), and the typical variable pairs with the largest correlation coefficient are retained. On this basis, the correlation between the blast furnace molten iron quality related variables and the typical variable pairs is further calculated, and the variables among the blast furnace molten iron quality related variables that are manipulated variables and closely related to the output variables are selected as input variables; the molten iron quality indicators are used as output variables.

[0034] Specifically, we focus on the correlation analysis between blast furnace molten iron quality related variables and molten iron quality indicators in the data set. The goal is to find the maximum correlation of the linear combination of two sets of variables, that is, to determine the linear combination of two sets of variables so that the correlation coefficient between the two linear combinations is maximized. This process can be mathematically rigorously expressed as a specific optimization problem: (1) In formula (1), the variables related to blast furnace molten iron quality are set as Hot metal quality index Given two sets of random data vectors; , The variables are and The linear combination coefficient vector of ; represents the covariance matrix of the variables; Representation variables The covariance matrix of Representation variables and The covariance matrix of .

[0035] In the specific implementation process, through complex mathematical operations, the optimal linear combination vector is found, so that the linear combination of the set of blast furnace molten iron quality related variables and the set of molten iron quality indicators is and The correlation coefficient between them reaches the maximum.

[0036] In this example, this method was used to conduct an in-depth analysis of 800 sets of experimental data collected. Through careful calculation and comparison, the potential correlation between blast furnace molten iron quality related variables and molten iron quality indicators was revealed, and typical variable pairs with a correlation coefficient of 0.5975 were successfully retained. These variable pairs provide key information for subsequent data analysis and model construction.

[0037] The correlation between the blast furnace molten iron quality related variables and the typical variable pairs is further calculated, and the variables that are easy to control and closely related to the output variables are selected as input variables. The molten iron quality index is used as the output variable. The input variable and output variable data constitute the data set For subsequent use.

[0038] Finally, the oxygen enrichment flow was confirmed , Cold air flow , hot air pressure , hot air temperature These four parameters are used as input variables of the data-driven model. This selection not only takes into account the correlation of the parameters, but also fully considers the controllability of the parameters and the actual situation of blast furnace operation. At the same time, through this characteristic variable selection operation, the redundant parts contained in the data are successfully removed, unnecessary calculations are reduced, and the complexity of subsequent models is reduced. This step provides strong support for the implementation of the subsequent blast furnace multivariate hot metal quality prediction and control method.

[0039] S23: Based on the input variables and output variables, considering the inherent timing relationship and delay effect of the blast furnace ironmaking process, determining the input and output variable combination; Specifically, based on the input variables and output variables, considering the inherent timing relationship and delay effect of the blast furnace ironmaking process, the input and output variable combination is determined to ensure that the subsequent soft measurement prediction model can accurately capture the dynamic relationship between the input and output variables, thereby improving the accuracy of the prediction of molten iron quality indicators and the actual application effect of the model.

[0040] In this case, the input and output timing coefficients of the process were determined through a detailed analysis of the sampling frequency of the molten iron quality. and (or called delay coefficient), are all set to 1, indicating that there is a direct and immediate relationship between the input variable and the output variable. Finally, the input and output variable combination of the model based on the time series relationship is specifically expressed as: (2) Based on the above time series relationship, the combination form of the model input and output variables was finally determined. This combination form fully considers the time series and time lag characteristics of the blast furnace smelting process, providing a solid foundation for the subsequent model training and prediction. By accurately capturing the nonlinear dynamic characteristics of the blast furnace smelting process, the quality of molten iron can be predicted more accurately, providing strong support for the optimization and control of the blast furnace ironmaking process.

[0041] S3: Based on the input and output variable combination, a back propagation neural network is applied to establish a soft measurement prediction model for online prediction of blast furnace molten iron quality indicators, including the following sub-steps: S31: constructing a soft measurement prediction model based on the input and output variable combination, wherein the soft measurement prediction model includes an input layer, multiple hidden layers and an output layer; Specifically, the number of neurons in the input layer is determined according to the input variables to ensure that the model can capture the key information of the blast furnace smelting process as comprehensively as possible. The number of layers and neurons in the hidden layer are selected through optimization to ensure that the model has sufficient capacity to capture complex nonlinear relationships and avoid overfitting.

[0042] In this example, three hidden layers are set, and each layer contains 16 neurons. The neuron activation function uses the ReLU function to enhance the nonlinear processing ability of the model and accelerate the training process. The number of neurons in the output layer is 2, corresponding to the predicted values ​​of the molten iron temperature and the molten iron silicon content in the molten iron quality index.

[0043] S32: Dataset using blast furnace smelting process The model is trained. During the training process, the weights and biases of the network are optimized. Through multiple iterations and adjustments, the performance of the model on the training data set is improved. At the same time, cross-validation and other techniques are used to evaluate the generalization ability of the model to ensure that it can maintain stable prediction accuracy on unseen data.

[0044] In this example, the trained soft sensor prediction model is applied to the online prediction of blast furnace molten iron quality. After receiving the input variable data of the blast furnace production process in real time, the model can quickly calculate the molten iron temperature. These forecast data provide timely and accurate data support for blast furnace operation adjustment, which helps to optimize the blast furnace smelting process and improve the quality of molten iron.

[0045] Table 1 below and Figure 3 The modeling and prediction results of the blast furnace molten iron temperature and molten iron silicon content quality indicators in this example show that the SC-MLP algorithm performs well in prediction accuracy, with low RMSE, MAE and MRE values. Figure 3 The displayed prediction effect diagram also further verified the excellent performance of the algorithm in set value tracking and control variable output, proving its effectiveness and reliability in blast furnace quality control.

[0046] Table 1: In order to comprehensively evaluate the performance of the prediction model, a series of standard evaluation indicators are used, including root mean square error (RMSE), mean absolute error (MAE), mean relative error (MRE) and hit rate (HR). Their calculation formulas are as follows: (3) (4) (5) (6) In formula (3)-(6), and Represent the actual value and the predicted value of the soft sensor prediction model respectively.

[0047] After rigorous evaluation, the soft sensor prediction model has shown excellent performance in all performance indicators. Specifically, in the prediction of molten iron temperature, the error of the soft sensor prediction model is controlled within ±10°C in 81.67%, which fully demonstrates the high accuracy of the model in temperature prediction. At the same time, in the prediction of molten iron silicon content concentration, the error of the soft sensor prediction model is kept within ±0.1 in 93.33%, which further proves the excellent performance of the model in silicon content prediction. In summary, the soft sensor prediction model in this example has significant advantages in prediction accuracy, which can fully meet the high requirements of blast furnace molten iron quality monitoring and provide strong data support for the optimization of blast furnace production.

[0048] S4: Based on the soft sensor prediction model, the feedback correction mechanism and the rolling optimization strategy are combined to control the blast furnace molten iron quality indicators. In the rolling optimization strategy process, the K-means clustering algorithm is introduced based on the historical operation data to divide the sub-modes, and the constrained optimization space is determined by the current working conditions, so as to ensure that the optimization process is always carried out within a safe range; Specifically, based on the soft measurement prediction model, in each control cycle, the future response is predicted in combination with the current control environment and corresponding control decisions are made.

[0049] In this example, the soft sensor prediction model is used as the prediction model, and the dynamic behavior characteristics of the blast furnace are accurately captured by training historical data. The model predicts the quality indicators of the blast furnace molten iron in each control cycle, and the prediction results based on the soft sensor prediction model provide support for control decisions.

[0050] Based on the control process, a feedback correction mechanism is introduced in the design of the predictive controller to improve the prediction accuracy of the soft measurement prediction model. The feedback correction mechanism is introduced as follows: by comparing the output measurement value of the actual object at the tth moment with the predicted output of the soft measurement prediction model, an error value is calculated, and the error value is used to adjust the output of the soft measurement prediction model at the t+1th moment.

[0051] In this example, The output measurement value of the actual object at the moment The prediction output of the soft sensor prediction model Compare and calculate the error, that is: (7) Then, use this error value to Time next Step prediction output Feedback adjustment is performed to calculate the output prediction value of the compensated soft sensor prediction model: (8) The rolling optimization strategy is as follows: in each sampling period, the future response of the controlled object is estimated using the soft measurement prediction model, and a real-time optimization performance indicator is constructed based on this: (9) In formula (9), the prediction time domain , control time domain , control weighting coefficient and output weighting coefficients are the unit diagonal matrices respectively. For input variables Time has come The value of change at a moment.

[0052] In this example, this real-time optimization index is solved by numerical optimization or swarm intelligence optimization method to generate the optimal control sequence. Then, the first control quantity of the sequence is applied in real time to the running model predictive control system (such as Figure 4 ), thereby achieving real-time optimization control.

[0053] Based on the rolling optimization strategy and the blast furnace historical data, the K-means clustering algorithm is innovatively introduced in the rolling optimization strategy to divide the blast furnace operating condition sub-modes. The cluster to which the current data point belongs is determined by calculating the Euclidean distance, and finally the sub-mode label is obtained and the constraint optimization space is determined for the current sub-mode, thereby ensuring that the optimization process is always carried out within a safe range. In this example, the K-means algorithm is used to cluster the input and output variables (i.e., molten iron temperature, molten iron silicon content, oxygen-enriched flow, cold air flow, hot air pressure, and hot air temperature) at the current moment in the historical output data, as well as the input variables at the previous moment (to avoid duplication, they are recorded as: oxygen-enriched flow 2, cold air flow 2, hot air pressure 2, and hot air temperature 2). First, randomly select k initial clustering centers. , then for each data point , calculate its distance from each cluster center, and assign the data point to the nearest cluster center. Using Euclidean distance as a metric, the clustering calculation formula is: (10) After that, recalculate the cluster center of each cluster, that is, calculate the number of nodes assigned to each cluster. The mean of all data points of : (11) In formula (11), is assigned to the cluster The number of data points, Belongs to the cluster The set of all data points.

[0054] The above two calculation processes are performed alternately until the cluster center no longer changes significantly or the maximum number of iterations is reached. Throughout the process, the goal of the K-means algorithm is to minimize the following objective function: (12) In formula (12), Represents each cluster The total distance from the data points in to the center of their cluster. Figure 5 The clustering results of blast furnace historical data are shown. Figure 6 The distribution of each control variable between two under different sub-modal labels is shown.

[0055] In each rolling optimization strategy, the current sub-mode is determined based on the current operating conditions and the input variables at the previous moment, and optimized within its corresponding small constraint space. This process prevents the non-analytical solution in the optimization process from exceeding the distribution of historical data, thereby reducing the risk of blast furnace operation and ensuring that the optimization process is limited to a safe range. Through this method, the safety and stability of blast furnace operation are improved. Through the implementation of the above steps, the soft sensor prediction model is successfully combined with the rolling optimization strategy and the sub-mode partitioning constraint optimization, such as Figure 2 As shown in the figure. The system significantly enhances the control system's dynamic and precise control capabilities in controlling the molten iron temperature and silicon content of the blast furnace. This lays a solid foundation for ensuring the safe and stable operation of the blast furnace under complex working conditions.

[0056] In this example, experimental data recorded in the actual production database of a blast furnace in a certain year and month of an ironmaking plant in South China were selected for detailed study. These data cover the key links of blast furnace production, including input parameters such as oxygen enrichment flow, cold air flow, hot air pressure, and hot air temperature, as well as output quality indicators such as molten iron temperature and molten iron silicon content. First, these raw data were thoroughly cleaned, and the characteristic variables were carefully selected through correlation analysis, and then the input and output variable pairing based on time sequence was determined.

[0057] In order to achieve real-time prediction of blast furnace molten iron quality indicators, a back propagation neural network was used to establish a soft sensor prediction model, which incorporates a back propagation algorithm to optimize network weights and biases. In particular, the model's architecture design fully considers the complexity and nonlinear characteristics of the blast furnace smelting process, ensuring the effective flow of information and the accuracy of processing. Figure 3 The modeling effect and prediction effect diagram of the soft sensor prediction model established based on the above data on blast furnace data are demonstrated.

[0058] In the tracking and control of the set values ​​of the silicon content and temperature of the molten iron, the K-means clustering algorithm was innovatively introduced to perform detailed sub-modal division of the blast furnace's operating status. This process not only helped determine the constrained optimization space, but also effectively limited the optimization process within a safe range, greatly reducing the risk of blast furnace operation.

[0059] Finally, under the framework of model predictive control, the rolling optimization strategy, feedback correction mechanism and submodal partitioning constraint optimization were combined to achieve dynamic and precise control of blast furnace molten iron temperature and molten iron silicon content. This comprehensive control strategy not only enhances the robustness of the system, but also ensures that the blast furnace can operate safely and stably under complex and changeable working conditions, providing solid technical support for the continuous optimization of blast furnace production and the continuous improvement of molten iron quality.

[0060] A comparative test was carried out to divide the sub-modes of the blast furnace operating state using the K-means clustering algorithm based on the blast furnace historical data, and to verify the effect of determining the constraint optimization space according to the current operating conditions in the control. Figure 7 A simulation effect diagram of setting value tracking effect and control variable output using constraint-added optimization predictive control according to an exemplary embodiment is shown; Figure 8 A schematic diagram showing the change of sub-modal labels during the prediction control process using constraint-added optimization; Fig. 9 The simulation effect diagram of the set value tracking effect and the control variable output using the unconstrained optimization predictive control is shown. In order to quantitatively evaluate the set value tracking control effect of the molten iron silicon content and molten iron temperature, the root mean square error is used as the key evaluation index. The evaluation index results are shown in Table 2.

[0061] Table 2: Through comparative experiments, the effect of adding constrained optimization is further improved, and the RMSE of molten iron temperature is reduced from 1.6435℃ to 1.2028℃, while the RMSE of molten iron silicon content is reduced from 0.0097 to 0.0082. The effectiveness and reliability of the blast furnace multivariate molten iron quality control method based on submodal partitioning constrained optimization proposed by the present invention are intuitively demonstrated.

[0062] Corresponding to the aforementioned embodiment of the blast furnace multi-element molten iron quality control method based on sub-modal partitioning constraint optimization, the present application also provides an embodiment of the blast furnace multi-element molten iron quality control device based on sub-modal partitioning constraint optimization.

[0063] Fig.10 1 is a block diagram of a blast furnace multivariate molten iron quality control device based on submodal partitioning constraint optimization according to an exemplary embodiment. Fig.10 , the device comprises: Data acquisition module 1, used to obtain the operation data in the blast furnace production process to construct the original data set, including blast furnace molten iron quality related variables and molten iron quality indicators; Data processing module 2 is used to clean the original data set, and then use the correlation analysis method to select the input variables and output variables of the model, and determine the input and output variable combination of the nonlinear dynamic characteristics in the blast furnace smelting process based on the time series relationship between the input and output variable data; Model building module 3, used for establishing a soft measurement prediction model based on the input and output variable combination by applying a back propagation neural network, so as to perform online prediction of the blast furnace molten iron quality index; The control module 4 is used to control the quality of molten iron in the blast furnace based on the soft measurement prediction model, combined with the feedback correction mechanism and the rolling optimization strategy. In the rolling optimization strategy process, the K-means clustering algorithm is introduced based on the historical operation data to divide the sub-modes, and the constrained optimization space is determined by the current operating conditions, so as to ensure that the optimization process is always carried out within a safe range.

[0064] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0065] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0066] Correspondingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the blast furnace multi-element molten iron quality control method based on sub-modal partitioning constraint optimization as mentioned above.

[0067] Correspondingly, the present application also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the above-mentioned blast furnace multivariate molten iron quality control method based on sub-modal partitioning constraint optimization.

[0068] The specific implementation methods described above have described the technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for controlling the quality of blast furnace multivariate molten iron based on submodal partitioning constraint optimization, characterized in that: include: The operating data of the blast furnace production process is obtained to construct the original data set, which includes variables related to blast furnace molten iron quality and molten iron quality indicators; The original data set is cleaned, and then the input variables and output variables of the model are selected by using a correlation analysis method, and the input and output variable combinations of the nonlinear dynamic characteristics in the blast furnace smelting process are determined based on the time series relationship between the input and output variable data; Based on the input and output variable combination, a back propagation neural network is applied to establish a soft measurement prediction model for online prediction of blast furnace molten iron quality indicators; Based on the soft sensor prediction model, the quality of molten iron in the blast furnace is controlled in combination with the feedback correction mechanism and the rolling optimization strategy. In the rolling optimization strategy process, the K-means clustering algorithm is introduced based on historical operating data to divide the sub-modes, and the constrained optimization space is determined by the current operating conditions, thereby ensuring that the optimization process is always carried out within a safe range.

2. The method according to claim 1, characterized in that The operating data of the blast furnace production process is obtained to construct the original data set, which contains blast furnace molten iron quality related variables and molten iron quality indicators, including: Obtain relevant production parameter process data from the blast furnace's historical operation database; Based on the relevant production parameter process data, the data are divided into blast furnace molten iron quality related variables and molten iron quality indicators according to the physical position of the variables in the blast furnace production process and their functional roles.

3. The method according to claim 1, characterized in that The original data set is cleaned, and then the input variables and output variables of the model are selected by using the correlation analysis method, and the input and output variable combinations of the nonlinear dynamic characteristics in the blast furnace smelting process are determined based on the time series relationship between the input and output variable data, including: Performing data cleaning on the original data set; Based on the cleaned data set, the correlation analysis method is used to find the best linear combination, so that the correlation coefficient between the blast furnace molten iron quality related variables and the molten iron quality indicators in the data set is maximized, and the typical variable pairs with the largest correlation coefficient are retained. On this basis, the correlation between the blast furnace molten iron quality related variables and the typical variable pairs is further calculated, and the variables that are manipulated variables and closely related to the output variables in the blast furnace molten iron quality related variables are selected as input variables; the molten iron quality indicators are used as output variables; Based on the input variables and output variables, the combination of input and output variables is determined taking into account the inherent timing relationship and delay effect of the blast furnace ironmaking process.

4. The method according to claim 1, characterized in that: Based on the input and output variable combination, a soft sensor prediction model is established by applying a back propagation neural network, including: Based on the input and output variable combination, a soft measurement prediction model is constructed, wherein the soft measurement prediction model includes an input layer, a plurality of hidden layers and an output layer, the number of neurons in the input layer is determined according to the input variables, and the number of layers and neurons in the hidden layer are selected by optimization; Using the blast furnace smelting process dataset The model is trained. During the training process, the weights and biases of the network are optimized. Through multiple iterations and adjustments, the performance of the model on the training data set is improved. At the same time, the generalization ability of the model is evaluated to ensure that it can maintain stable prediction accuracy on unseen data.

5. The method according to claim 1, characterized in that The feedback correction mechanism is introduced as follows: by comparing the output measurement value of the actual object at the tth moment with the predicted output of the soft measurement prediction model, an error value is calculated, and the error value is used to adjust the output of the soft measurement prediction model at the t+1th moment.

6. The method according to claim 1, characterized in that The rolling optimization strategy is specifically as follows: in each sampling period, the future response of the controlled object is estimated using a soft measurement prediction model, and a real-time optimization performance indicator is constructed based on this, and then the optimal control sequence is solved, and the first control quantity of the optimal control sequence is applied to the model predictive control system, thereby achieving real-time optimization control.

7. The method according to claim 1, characterized in that In the process of rolling optimization strategy, K-means clustering algorithm is introduced based on historical operation data to divide the sub-modes, and the constraint optimization space is determined by the current working conditions, including: Based on the rolling optimization strategy and the historical data of the blast furnace, the K-means clustering algorithm is introduced to divide the sub-modes of the blast furnace operation conditions. Specifically, the input and output variables at the current moment and the previous moment are clustered, and the cluster to which the data point belongs is determined by calculating the Euclidean distance. Finally, the sub-mode label is obtained and the constraint optimization space is determined for the current sub-mode.

8. A blast furnace multivariate molten iron quality control device based on submodal partitioning constraint optimization, characterized in that: include: The data acquisition module is used to obtain the operating data in the blast furnace production process to construct the original data set, including blast furnace molten iron quality related variables and molten iron quality indicators; A data processing module is used to clean the original data set, and then use a correlation analysis method to select the input variables and output variables of the model, and determine the input and output variable combination of the nonlinear dynamic characteristics in the blast furnace smelting process based on the time series relationship between the input and output variable data; A model building module, for establishing a soft measurement prediction model based on the input and output variable combination and applying a back propagation neural network, for online prediction of blast furnace molten iron quality indicators; A control module is used to control the quality of molten iron in the blast furnace based on the soft sensor prediction model, combined with a feedback correction mechanism and a rolling optimization strategy. In the rolling optimization strategy process, a K-means clustering algorithm is introduced based on historical operating data to divide the sub-modes, and the constrained optimization space is determined by the current operating conditions, thereby ensuring that the optimization process is always carried out within a safe range.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images