Blast furnace silicon content prediction method based on gaussian mixture model and improved residual analysis

Abstract

The present application relates to the technical field of automation detection, and more particularly to a blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis. First, by integrating the silicon content, corresponding material parameters and environmental parameters in the database within a preset time, a first basic data set is constructed; then, using Gaussian mixture model and improved residual analysis, input abnormal data and output abnormal data are screened respectively to obtain second and third abnormal data sets, and by eliminating the data contained in the two abnormal data sets, a fourth high-quality data set is obtained; then an XGBoost regression prediction model is built, and the fourth high-quality data set is used to train the model, and finally a high-quality XGBoost regression prediction model is obtained. This model can predict the blast furnace silicon content in real time, and through the screening of abnormal data, the quality of the data set is improved, and the accuracy of the blast furnace silicon content prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of automated detection technology, and in particular to a method for predicting the silicon content of blast furnaces based on Gaussian mixture models and improved residual analysis. Background Technology

[0002] As a fundamental and crucial industry in industrial production, the steel industry relies heavily on blast furnace ironmaking as its prerequisite and core issue. Improving ironmaking efficiency is of great economic and environmental significance, helping steel companies achieve remarkable results in cost reduction, efficiency improvement, energy conservation, and emission reduction. In this industrial chain, real-time information on the silicon content in molten iron is crucial for blast furnace operators to assess the internal condition and pig iron quality. By understanding changes in silicon content in advance, operators can take accurate control measures when necessary, effectively improving and stabilizing the quality of molten iron.

[0003] Traditional methods for detecting silicon content involve sending molten iron from the furnace to a laboratory for analysis. However, this method suffers from significant time lag and offers limited guidance for frontline production. Furthermore, while relying on operators' professional experience and process knowledge for silicon content assessment offers real-time results, it suffers from drawbacks such as non-transferability, uncertainty, and ambiguity, and its judgment is heavily influenced by the operators' experience and knowledge. Due to the complex physicochemical reactions involved within the blast furnace, traditional formula fitting methods struggle to comprehensively describe and decouple the changes in silicon content.

[0004] To overcome the aforementioned problems, machine learning and other methods are introduced to build models for predicting silicon content, which possess the characteristics of theoretical rigor, learnability, and portability. This method can more comprehensively study the prediction of silicon content in blast furnace hot metal, not only for predicting silicon content but also laying the foundation for further research on a series of quality indicators, including silicon content. However, the massive amounts of data accumulated in the databases of steel enterprises contain a large number of low-quality and abnormal data. Using such data to train regression prediction models will contaminate the models to varying degrees, reducing their predictive effectiveness.

[0005] In summary, to address the current problems of poor real-time performance, poor portability, and contamination of regression prediction models with inferior blast furnace data, a blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis is proposed. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for predicting blast furnace silicon content based on Gaussian mixture model and improved residual analysis, thereby achieving accurate prediction of blast furnace silicon content.

[0007] This invention provides a method for predicting silicon content in blast furnaces based on Gaussian mixture models and improved residual analysis, comprising the following steps: S100. Integrate the silicon content data within a preset time period in the database, along with the material parameters and environmental parameters corresponding to the silicon content within that time period, as the initial dataset to obtain the first basic dataset. S200. Use a Gaussian mixture model to filter out abnormal input data from the first basic dataset to obtain the second abnormal dataset. S300. Use improved residual analysis to filter out outliers in the first basic dataset to obtain the third outlier dataset. S400. Based on the first basic dataset, remove the data contained in the second abnormal dataset and the data contained in the third abnormal dataset to obtain a fourth high-quality dataset. S500. Build an XGBoost regression prediction model and train it using the fourth high-quality dataset to obtain a high-quality XGBoost regression prediction model. The S600 uses a high-quality XGBoost regression prediction model to predict the silicon content of blast furnaces in real time.

[0008] Furthermore, in S100, the process of integrating the silicon content stored in the database by the steel enterprise for a certain period of time, along with the corresponding material and environmental parameters for that silicon content, as an initial dataset to form the first basic dataset includes: S101. Perform Spearman correlation analysis and grey relational analysis on the initial dataset to calculate the correlation and degree of correlation between silicon content and material parameters and environmental parameters. By setting a first correlation threshold and a second correlation threshold, input variables with Spearman correlation less than the first correlation threshold and input variables with grey relational correlation less than the second correlation threshold are removed. S102. Using the XGBoost-RFECV method, the initial dataset that has been filtered out by Spearman correlation analysis and grey relational analysis is optimized to obtain the first basic dataset.

[0009] Furthermore, in step S200, the step of using a Gaussian mixture model to filter input anomalies in the first basic dataset to obtain a second anomaly dataset includes: S201. Randomly generate initial parameters for multiple Gaussian distributions, wherein the initial parameters for each Gaussian distribution include mean, covariance matrix and weights; S202. Use the EM algorithm to iteratively process the Gaussian mixture model, repeatedly execute the E step and the M step, and update the initial parameters of each Gaussian distribution until the predetermined number of iterations is reached to obtain the trained Gaussian mixture model. S203. Use the trained Gaussian mixture model to evaluate the first basic dataset and obtain the Gaussian mixture score. Data in the first basic dataset with a Gaussian mixture score less than the evaluation threshold are used as the second abnormal dataset.

[0010] Furthermore, in step S300, the step of using improved residual analysis to filter out outlier data from the first basic dataset to obtain a third outlier dataset includes: S301. Using the idea of ​​K-fold cross-validation, the first basic dataset is divided into K mutually exclusive subsets; S302. Use the bootstrap method to resample the data of all subsets except the k-th subset multiple times, where k=1,2,3,...,K. Each resampling generates a new training set. For each training set, train an independent regression model and record the prediction results. Using the prediction results of multiple regression models, calculate the cognitive uncertainty and random uncertainty of each data in the generated training set, and then obtain the cognitive uncertainty and random uncertainty of all data in the first basic dataset. S303. Perform improved residual analysis on the first basic dataset calculated by the bootstrap method. Calculate the improved residual of all data in the first basic dataset based on the cognitive uncertainty and random uncertainty of all data in the first basic dataset. Take all abnormal data with improved residuals higher than the abnormal threshold as the third abnormal dataset.

[0011] Furthermore, in step S400, the step of removing data from the second and third abnormal datasets based on the first basic dataset to obtain a fourth high-quality dataset includes: The Gaussian mixture model and the improved residual analysis are designed in parallel. Based on the first basic dataset, the data contained in the second and third abnormal datasets are removed to obtain the fourth high-quality dataset.

[0012] Furthermore, in S500, the construction of the XGBoost regression prediction model includes: Select the preset maximum tree depth and number of decision trees.

[0013] Furthermore, in S600, the real-time prediction of blast furnace silicon content using a high-quality XGBoost regression prediction model includes: S601. Collect real-time blast furnace operation data under working conditions; wherein, the real-time blast furnace operation data includes material parameters and environmental parameters; S602. Input the real-time blast furnace operation data into the trained Gaussian mixture model for analysis; if the score of the real-time blast furnace operation data output by the trained Gaussian mixture model is lower than a preset threshold, the real-time blast furnace operation data is determined to be abnormal data, prediction is stopped and a warning is issued; if the score of the real-time blast furnace operation data output by the trained Gaussian mixture model is not lower than the preset threshold, the real-time blast furnace operation data is determined to be normal data, and the real-time blast furnace operation data is input into the high-quality XGBoost regression prediction model to predict the silicon content of the blast furnace in real time.

[0014] The present invention also provides an electronic device, the electronic device comprising: Processor and memory; The processor executes the steps of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis as described above by calling programs or instructions stored in memory.

[0015] The present invention also provides a computer-readable storage medium storing a program or instructions that cause a computer to perform the steps of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis as described in any of the preceding claims.

[0016] The embodiments of the present invention have the following technical effects: 1. Optimizing Data Quality and Reducing Prediction Errors: The databases of steel enterprises contain a large amount of low-quality and outlier data, which contaminates regression prediction models and reduces prediction effectiveness. By employing techniques such as Spearman correlation analysis, grey relational analysis, Gaussian mixture models, and improved residual analysis, outlier data was successfully eliminated, constructing a high-quality dataset. This data optimization method effectively improved the performance of the XGBoost regression prediction model, reduced prediction errors, and provided more reliable prediction results for blast furnace silicon content prediction.

[0017] 2. Improved Real-Time Performance and Portability: Traditional silicon content detection methods suffer from latency, while methods relying on operator experience lack portability. This invention introduces machine learning technology, particularly the XGBoost regression prediction model, to predict silicon content in real time. This method not only improves real-time performance but also demonstrates strong portability under different environments and conditions, enabling steel companies to more timely and accurately grasp the dynamic changes in blast furnace silicon content.

[0018] 3. Improving Prediction Accuracy through Comprehensive Application of Multiple Techniques: This invention comprehensively applies multiple techniques, including Spearman correlation analysis, grey relational analysis, Gaussian mixture model, improved residual analysis, and XGBoost regression prediction model, to form a comprehensive method for predicting silicon content in blast furnaces. This comprehensive method overcomes the difficulties of traditional methods in handling complex physicochemical reaction processes within blast furnaces, thus improving the accuracy of silicon content prediction. Through multi-level and multi-angle data processing and model building, it effectively solves the limitations of traditional methods, providing the steel industry with a more reliable and intelligent silicon content prediction solution. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] Figure 1 This is a schematic diagram of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis provided in an embodiment of the present invention. (See also...) Figure 1 Specifically, it includes: S100. Integrate the silicon content within a preset time period in the database, along with the corresponding material and environmental parameters for that time period, as the initial dataset to obtain the first basic dataset. For example: the silicon content and corresponding material and environmental parameters in the database of a steel company during the period from 0:00 on January 1, 2023 to 23:00 on October 31, 2023 are read as the initial dataset, and the initial dataset is integrated to obtain the first basic dataset.

[0023] S100 specifically includes: S101. First, the Spearman correlation coefficient is used to measure the nonlinear relationship between silicon content and material parameters and environmental parameters. The Spearman correlation coefficient is a nonparametric statistical method, which is particularly suitable for permutation data and is used to measure the monotonic relationship between two variables. Its calculation results can provide information on the direction and strength of the correlation.

[0024] Next, grey relational analysis was used to further evaluate the correlation between silicon content and each input parameter. Grey relational analysis is a multivariate correlation analysis method that can be used to determine the relative strength of the correlation between different variables. By setting a first correlation threshold of 0.1 and a second correlation threshold of 0.6, input variables with Spearman correlation less than the first correlation threshold and input variables with grey relational degree less than the second correlation threshold were removed. The goal of this screening process is to remove input variables with weak or no correlation with silicon content, ensuring that the final feature set contains highly correlated information, providing a more accurate and high-quality data foundation for subsequent modeling, and improving the reliability of the silicon content prediction model.

[0025] S102. Using the RFECV algorithm with XGBoost as the basic regressor, the algorithm selects the subset of features that contribute most to the silicon content prediction performance from the 22-dimensional input feature space, which serves as the first basic dataset. This algorithm continuously eliminates features with minimal impact on model performance through a combination of recursive feature elimination and cross-validation, ultimately forming the first basic dataset. The first basic dataset includes material parameters and environmental parameters as input variables in subsequent steps, and silicon content as the output variable in subsequent steps.

[0026] S200. Use a Gaussian mixture model to filter out abnormal input data from the first basic dataset to obtain the second abnormal dataset. Specifically, it includes: S201. First, initialize the Gaussian mixture model. Initialize a Gaussian mixture model with 4 Gaussian distributions. The dimension of the Gaussian distribution is set to 22. At the same time, the initial parameters are randomly generated. Use all input feature data of the first basic dataset to train the Gaussian mixture model.

[0027] In this Gaussian mixture model, the observed data is a mixture of multiple Gaussian distributions. Each Gaussian distribution is determined by its mean. Covariance matrix and weight The distribution is described by the mean, which represents the center of the distribution; the covariance matrix describes the shape of the distribution; and the weights represent the relative importance of each distribution in the mixture.

[0028] S202. Since the parameters of the Gaussian Mixture Model (GMM) are initially randomly generated, it lacks the ability to reflect the data distribution characteristics within the dataset. Therefore, the EM algorithm is needed to train the GMM until convergence, while simultaneously ensuring that the GMM has the ability to calculate the probability density of any 22-dimensional data. The formula is:

[0029] Where n represents the nth data group in the first basic dataset, A represents A Gaussian distributions, and a represents the ath Gaussian distribution. It is the input feature of the nth data group in the first basic dataset. These are the weights of the a-th Gaussian distribution. It is the mean of the a-th Gaussian distribution. It is the covariance matrix of the a-th Gaussian distribution. Let A represent the density function of the a-th multivariate Gaussian distribution, where A is taken as 4 in the formula.

[0030] Subsequently, the Gaussian mixture model is trained using the EM algorithm. In the E step, the posterior probability of each data point in the first base dataset belonging to each Gaussian distribution is calculated. The posterior probability is the probability that the nth data point belongs to the a-th distribution. The formula for the posterior probability is as follows:

[0031] in, It is the posterior probability. It is the weight of the b-th Gaussian distribution. It is the mean of the b-th Gaussian distribution. It is the covariance matrix of the b-th Gaussian distribution. Let A represent the density function of the b-th multivariate Gaussian distribution, where A is taken as 4 in the formula.

[0032] For any 22-dimensional input data in the first basic dataset, its Gaussian mixture posterior probability is a value between 0 and 1, which represents the probability that the data belongs to the a-th distribution. The larger the value, the more likely the data is to truly belong to that distribution.

[0033] In step M, the parameters of each Gaussian distribution will be updated using the posterior probabilities calculated in step E.

[0034]

[0035]

[0036] Where N is the total data size of the first basic dataset. These are the weights of the updated Gaussian distribution. It is the mean of the updated Gaussian distribution. It is the covariance matrix of the updated Gaussian distribution.

[0037] As the EM algorithm iterates over the Gaussian mixture model, the mean, covariance matrix, and weight parameters of the four Gaussian distributions will gradually match the distribution of the data in the first basic dataset.

[0038] At the start of training using the EM algorithm, the maximum number of iterations is set to 100. If the number of iterations reaches 100, the iteration will stop and be considered complete. At the same time, before reaching the maximum number of iterations, the changes in the Gaussian mixture model parameters will be checked after each iteration. If the changes in parameters are less than a predefined threshold, the Gaussian mixture model is also considered to have converged and the iteration is considered complete.

[0039] S203. After several rounds of training, the Gaussian Mixture Model (GMM) converges. At this point, any data of the same dimension can be input into the converged GMM to calculate a GMM score. A higher score indicates that the sample conforms more closely to the model, while a lower score indicates a higher degree of data anomaly. The trained GMM is used to calculate the GMM score of all input data in the first base dataset, and an evaluation threshold of 0.05 is set. This means that the bottom 5% of data with the lowest GMM scores will be considered the second abnormal dataset. The formula for calculating the GMM score is as follows:

[0040] in, It is a Gaussian mixture fraction. It is the covariance matrix of the a-th distribution, and A is taken as 4 in the formula.

[0041] S300. Use improved residual analysis to filter out outlier data from the first basic dataset to obtain the third outlier dataset.

[0042] Specifically, it includes: S301. Using the idea of ​​K-fold cross-validation, where K is 5, the first basic dataset is divided into 5 mutually exclusive subsets, which are denoted as datasets [1], [2], [3], [4], [5].

[0043] S302. Use the bootstrap method to train the regression model on all or part of the data in [2], [3], [4], [5], and then validate the data in dataset [1] to obtain the out-of-sample prediction of the data in [1]. The same method can be used to obtain the out-of-sample prediction of dataset [2], [3], [4], [5].

[0044] The following explanation uses dataset [1] as an example. Data from datasets [2], [3], [4], and [5] are used for five bootstrap resampling operations. Each resampling generates a new training set. For each training set, an independent XGBoost regression prediction model is trained, and the prediction results are recorded. Here, we use... This represents the prediction results of the five models on the dataset [1], and thus the cognitive uncertainty and random uncertainty of each data point in the dataset [1] are obtained. The formula for cognitive uncertainty is:

[0045] In the above formula, It is cognitive uncertainty. It is the prediction result of the regressor obtained by training the nth group of data in the first basic dataset with several bootstrap resampling.

[0046] The formula for random uncertainty is:

[0047] In the above formula, It is random uncertainty. It is the input data for the nth group of data in the first basic dataset. This is the output data of the nth group of data in the first basic dataset. It is the prediction result of the regressor obtained by training the nth group of data in the first basic dataset with several bootstrap resampling.

[0048] The formula for the average value in residual analysis is:

[0049] Obtain the average residuals of the five independent regression models; in the above formula, It is the average value of the residual analysis. It is the prediction result of the regressor obtained by training the nth group of data in the first basic dataset with the cth bootstrap resampling, where C is the number of independent regression models, and C is 5 here.

[0050] Furthermore, the improved residual analysis was calculated, and the formula is as follows:

[0051] In the above formula, It is the input data for the nth group of data in the first basic dataset. This is the output data of the nth group of data in the first basic dataset. It is cognitive uncertainty. It is random and uncertain.

[0052] The improved residual analysis results for all data within the first basic dataset can be obtained using the above method. Based on the improved residual analysis results, the improved residual value is calculated for each data point using the improved residual analysis method, and an outlier threshold of 0.05 is set, meaning that data with improved residual values ​​higher than 5% will be identified as the third outlier dataset.

[0053] S400. Based on the first basic dataset, remove the data contained in the second abnormal dataset and the data contained in the third abnormal dataset to obtain the fourth high-quality dataset.

[0054] Specifically, it includes: When removing outliers, if a record in the data contains outliers in its input or output, then that entire record will be removed; that is, both the input and output will be removed simultaneously. This process combines two data filtering methods, comprehensively considering different anomaly detection perspectives, thus improving the comprehensiveness and accuracy of data filtering and enabling more thorough identification and removal of outliers.

[0055] S500. Build an XGBoost regression prediction model and train it using the fourth high-quality dataset to obtain a high-quality XGBoost regression prediction model. Specifically, this includes: using an XGBoost regression prediction model, setting the maximum tree depth to 3, the number of decision trees to 100, the learning rate to 0.1, and training it using the fourth high-quality dataset to obtain the final high-quality XGBoost regression prediction model.

[0056] The S600 uses a high-quality XGBoost regression prediction model to predict the silicon content of blast furnaces in real time.

[0057] Specifically, it includes: S601. First, collect real-time data of blast furnace operation in the working environment. The real-time data of blast furnace operation includes material parameters and environmental parameters. Then, organize the real-time data of blast furnace operation to obtain 22-dimensional data as real-time input. S602. Input the real-time input into the trained Gaussian mixture model for analysis and obtain the Gaussian mixture score. If the score is lower than the threshold of -240, it is considered abnormal data, and the prediction is stopped and a warning is issued. If the real-time data of blast furnace operation is determined to be normal data, the real-time data of blast furnace operation is input into the high-quality XGBoost regression prediction model to predict the silicon content of blast furnace in real time.

[0058] Figure 2 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. For example... Figure 2 As shown, the electronic device 500 includes one or more processors 501 and memory 502.

[0059] The processor 501 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 500 to perform desired functions.

[0060] The memory 502 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 501 may execute the program instructions to implement the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis in any embodiment of this application described above, and / or other desired functions. Various contents such as initial extrinsic parameters and thresholds may also be stored in the computer-readable storage medium.

[0061] In one example, the electronic device 500 may further include an input device 503 and an output device 504, these components being interconnected via a bus system and / or other forms of connection mechanisms (not shown). The input device 503 may include, for example, a keyboard, a mouse, etc. The output device 504 may output various information to the outside, including warning messages, braking force, etc. The output device 504 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0062] Of course, for the sake of simplicity, Figure 2 Only some of the components of the electronic device 500 relevant to this application are shown in this illustration; components such as buses, input / output interfaces, etc., are omitted. In addition, the electronic device 500 may include any other suitable components depending on the specific application.

[0063] In addition to the methods and devices described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis provided in any embodiment of this application.

[0064] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0065] Furthermore, embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis provided in any embodiment of this application.

[0066] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0067] It should be noted that the terminology used in this invention is for describing specific embodiments only and is not intended to limit the scope of this application. As shown in this specification, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may include the plural. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element.

[0068] It should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Unless otherwise expressly specified and limited, the terms "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood according to the specific circumstances.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting silicon content in blast furnaces based on Gaussian mixture model and improved residual analysis, characterized in that, Includes the following steps: S100. Integrate the silicon content data within a preset time period in the database, along with the material parameters and environmental parameters corresponding to the silicon content within that preset time period, as an initial dataset to obtain the first basic dataset. S200. Use a Gaussian mixture model to filter out abnormal input data from the first basic dataset to obtain the second abnormal dataset. S300. Use improved residual analysis to filter out outliers in the first basic dataset to obtain the third outlier dataset. S400. Based on the first basic dataset, remove the data contained in the second abnormal dataset and the data contained in the third abnormal dataset to obtain a fourth high-quality dataset. S500. Build an XGBoost regression prediction model and train it using the fourth high-quality dataset to obtain a high-quality XGBoost regression prediction model. S600 uses a high-quality XGBoost regression prediction model to predict the silicon content of blast furnace in real time. In step S200, the step of using a Gaussian mixture model to filter input anomalies in the first basic dataset to obtain a second anomaly dataset includes: S201. Randomly generate initial parameters for multiple Gaussian distributions, wherein the initial parameters for each Gaussian distribution include mean, covariance matrix and weights; S202. Use the EM algorithm to iteratively process the Gaussian mixture model, repeatedly execute the E step and the M step, and update the initial parameters of each Gaussian distribution until the predetermined number of iterations is reached to obtain the trained Gaussian mixture model. S203. Use the trained Gaussian mixture model to evaluate the first basic dataset and obtain the Gaussian mixture score. Data in the first basic dataset with a Gaussian mixture score less than the evaluation threshold are used as the second abnormal dataset. In step S300, the step of using improved residual analysis to filter out outlier data from the first basic dataset to obtain a third outlier dataset includes: S301. Using the idea of ​​K-fold cross-validation, the first basic dataset is divided into a mutually exclusive subsets; S302. Use the bootstrap method to resample the data of all subsets except the k-th subset multiple times. Each resampling generates a new training set. For each training set, train an independent regression model and record the prediction results. Using the prediction results of multiple regression models, calculate the cognitive uncertainty and random uncertainty of each data in the generated training set, and then obtain the cognitive uncertainty and random uncertainty of all data in the first basic dataset. S303. Perform improved residual analysis on the first basic dataset after the bootstrap method. Calculate the improved residual of all data in the first basic dataset based on the cognitive uncertainty and random uncertainty of all data in the first basic dataset. Take all abnormal data with improved residuals higher than the abnormality threshold as the third abnormal dataset. In S600, the real-time prediction of blast furnace silicon content using a high-quality XGBoost regression prediction model includes: S601. Collect real-time blast furnace operation data under working conditions; wherein, the real-time blast furnace operation data includes material parameters and environmental parameters; S602. Input the real-time blast furnace operation data into the trained Gaussian mixture model for analysis; if the score of the real-time blast furnace operation data output by the trained Gaussian mixture model is lower than a preset threshold, the real-time blast furnace operation data is determined to be abnormal data, prediction is stopped and a warning is issued; if the score of the real-time blast furnace operation data output by the trained Gaussian mixture model is not lower than the preset threshold, the real-time blast furnace operation data is determined to be normal data, and the real-time blast furnace operation data is input into the high-quality XGBoost regression prediction model to predict the silicon content of the blast furnace in real time.

2. The method for predicting blast furnace silicon content based on Gaussian mixture model and improved residual analysis according to claim 1, characterized in that... The S100 includes: S101. Perform Spearman correlation analysis and grey relational analysis on the initial dataset to calculate the correlation and degree of correlation between silicon content and material parameters and environmental parameters. By setting a first correlation threshold and a second correlation threshold, input variables with Spearman correlation less than the first correlation threshold and input variables with grey relational correlation less than the second correlation threshold are removed. S102. Using the XGBoost-RFECV method, the initial dataset that has been filtered out by Spearman correlation analysis and grey relational analysis is optimized to obtain the first basic dataset.

3. The method for predicting blast furnace silicon content based on Gaussian mixture model and improved residual analysis according to claim 1, characterized in that... In step S400, the step of removing data from the second and third abnormal datasets based on the first basic dataset to obtain a fourth high-quality dataset includes: The Gaussian mixture model and the improved residual analysis are designed in parallel. Based on the first basic dataset, the data contained in the second and third abnormal datasets are removed to obtain the fourth high-quality dataset.

4. The method for predicting blast furnace silicon content based on Gaussian mixture model and improved residual analysis according to claim 1, characterized in that... In S500, the construction of the XGBoost regression prediction model includes: Select the preset maximum tree depth and number of decision trees.

5. An electronic device, characterized in that, The electronic device includes: Processor and memory; The processor invokes the program or instructions stored in the memory to execute the steps of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program or instructions that cause a computer to perform the steps of the blast furnace silicon content prediction method based on Gaussian mixture model and improved residual analysis as described in any one of claims 1 to 4.

Images