Method for constructing blast furnace stability index system based on process mechanism and data-driven

Abstract

The present invention provides a method for constructing a blast furnace stability index system based on process mechanism and data-driven, and relates to the technical field of blast furnace ironmaking production. First, the data is preprocessed to eliminate mutations and spikes in the original data, so that the data tends to be smooth, and unnecessary redundant parameters are eliminated. Secondly, a new feature cycle selection method is used to establish a blast furnace stability index system, and the importance score of each feature is obtained based on the prediction error change analysis of the machine learning model, and the features with low importance are cyclically deleted based on the designed judgment rules and the actual needs of field experts, and finally the construction of the blast furnace stability index system is realized. The present invention can automatically construct a blast furnace stability index system based on field needs, can meet the monitoring needs of field experts for blast furnace conditions, help field operators better monitor and control the operating status of the blast furnace, and realize continuous improvement and optimization of the blast furnace production process.

Description

Technical Field

[0001] The present invention relates to the technical field of blast furnace ironmaking production, and in particular to a method for constructing a blast furnace stability index system based on process mechanism and data driving. Background Art

[0002] Blast furnace conditions refer to the operating state of the blast furnace during the ironmaking process, including the physical and chemical environment inside the blast furnace, the distribution and flow of the charge, the temperature and pressure distribution, the efficiency of the combustion and reduction reactions, and the physical state of the furnace body. The stable operation of the blast furnace is crucial to ensuring production efficiency, product quality, and reducing costs.

[0003] By analyzing historical data, establishing a stability index system for the blast furnace operation process, real-time monitoring of the blast furnace operation status and the changing trend of key parameters, timely discovering abnormal situations and taking measures to adjust and optimize production strategies, it is an important means to improve production efficiency and product quality, reduce production costs and energy consumption, and ensure the stability of the blast furnace production process. Therefore, establishing a blast furnace stability index system is of great significance for blast furnace operation, production management and product quality. Patent CN117094576A discloses a method for quantitatively evaluating the stability of blast furnace conditions. Under the conditions of given blast furnace characteristics, a quantitative evaluation method for evaluating the stability of blast furnace conditions is provided, but the technical method is very dependent on the experience of on-site experts when selecting the core parameters that can characterize the blast furnace conditions, and the parameters selected manually often have certain limitations. Therefore, how to automatically select the core parameters that affect the blast furnace conditions based on process mechanism and production data and construct a blast furnace stability index system is a key problem to be solved. Summary of the invention

[0004] The technical problem to be solved by the present invention is to address the deficiencies of the above-mentioned prior art and provide a method for constructing a blast furnace stability index system based on process mechanism and data-driven, so as to improve the accuracy of stability prediction of the blast furnace operation process and help on-site operators better monitor and control the operating status of the blast furnace.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for constructing a blast furnace stability index system based on process mechanism and data-driven includes the following steps:

[0007] Step 1: Obtain an initial dataset for the blast furnace ironmaking process ,in, is the feature vector composed of all collected items of the i-th sample, is the molten iron output of the i-th sample, n is the number of samples, m is the sample dimension, represents m-dimensional real number space;

[0008] Step 2: For the initial dataset in step 1 Perform data preprocessing, first for the initial data set All the collected items are classified and preliminarily selected to obtain the data set Then, the data set Perform data filtering and normalization to obtain the data set , and the dataset As the current sample set;

[0009] Step 3: For the current sample set, calculate the importance score of each feature in the current sample set;

[0010] Step 4: Based on the importance scores of the current sample set and its features, a blast furnace stability index system is constructed cyclically;

[0011] Step 5: Based on the stability index system of blast furnace process data obtained in step 4, a blast furnace stability evaluation model is constructed using appropriate machine learning modeling methods.

[0012] Furthermore, in step 2, the specific process of parameter classification and preliminary selection is as follows:

[0013] Step 2.1.1: According to the classification standard of blast furnace production data, all collected items are divided into four categories: condition parameters, operation parameters, process parameters and result parameters;

[0014] Step 2.1.2: According to the process mechanism, the required features are determined from the above four categories of parameter features, and the redundant features are eliminated. The remaining features are used as candidate features of the blast furnace stability index system to obtain the data set. , n is the number of samples, is the sample dimension after removing redundant features.

[0015] Furthermore, the data filtering process in step 2 uses a moving average filtering method to filter the data set. The original data in the filter is filtered to eliminate the mutation and peak phenomenon in the original data, making the data smooth and obtaining the data set. ; The filtering formula is as follows:

[0016] ;

[0017] Among them, k represents the window size of the moving average filter, represents the moving average, n is the number of samples, represents the feature column vector composed of the jth feature, .

[0018] Furthermore, the normalization process in step 2 uses MinMax to filter the data set after the data is processed. Do normalization and get the data set ; The normalization formula is as follows:

[0019] ;

[0020] in, represents the jth feature of the i-th sample, ; represents the minimum value of the jth feature, represents the maximum value of the jth feature, for Normalized value.

[0021] Furthermore, the specific steps of step 3 are:

[0022] Step 3.1: Split the current sample set into a training set and test set ;

[0023] Step 3.2: Use the training set Train a machine learning-based prediction model , and calculate the model In the test set The root mean square error on ;

[0024] Step 3.3: For each feature column vector in the current sample set , calculate the feature in the model The error variation on , get the error change of all features ;

[0025] Step 3.4: Normalize the error variation of all features and record it as ,in ; The normalized result is then used as the importance score of each feature.

[0026] Furthermore, in step 3.3, the error variation range The calculation method is:

[0027] Step 3.3.1: Generate h groups of Gaussian noise with different intensities ,in , and For the specified noise mean and variance; Gaussian noise with different means and the same variance is added to the test set in sequence Each characteristic column vector of In the new data set ;

[0028] Step 3.3.2: Dataset Represents the test set The characteristic column vector of Add Gaussian noise to generate a new dataset and calculate the trained machine learning model on the dataset The root mean square error on ;

[0029] Step 3.3.3: Repeat steps 3.3.1 and 3.2.2 t times and take the mean of the results to obtain the feature vector The error variation is:

[0030] ;

[0031] in, ; .

[0032] Furthermore, the step 4 specifically includes the following steps:

[0033] Step 4.1: For all features of the current sample set, find the feature column vector with the lowest feature importance score , the feature Remove from the current sample set, where And get a new data set ;

[0034] Step 4.2: The dataset obtained in step 4.1 As the current sample set, and split into training sets and test set ;

[0035] Step 4.3: Use the training set Training the model , computational model In the test set The prediction effect is the coefficient of determination , and record the feature combination of the current sample set; The calculation formula is as follows:

[0036] ;

[0037] in, represents the molten iron output of the i-th sample, represents the predicted value of molten iron output of the i-th sample, represents the mean of molten iron production of all samples;

[0038] Step 4.4: Determine whether the stopping criteria are met. If the stopping criteria are met, the final blast furnace stability index system will be output and step 5 will be entered; otherwise, return to step 3, continue to calculate the feature importance score for the new current sample set and delete the features with low importance.

[0039] Furthermore, the specific method of step 4.4 is:

[0040] Step 4.4.1: Get the coefficient of determination of the model calculated in step 4.3 Value, that is, the model removes features The prediction effect on the subsequent data set is recorded as , and remove the feature The decision coefficient of the previous model value For comparison; if , then return to step 3; otherwise, go to step 4.4.2;

[0041] Step 4.4.2: Determine the features to be removed The coefficient of determination after Is it below the threshold of the coefficient of determination set by the on-site experts? .like , then return to step 3; otherwise, it indicates that the feature is removed. The subsequent feature combination can no longer meet the accuracy requirements of the field experts for the stability evaluation model, so the feature The previous feature combination is regarded as the final blast furnace stability index system, and the cycle ends and enters step 5.

[0042] The beneficial effect produced by adopting the above technical scheme is that: the method for constructing a blast furnace index system based on process mechanism and data-driven provided by the present invention first pre-processes the data, eliminates mutations and spikes in the original data, makes the data smoother, and eliminates some unnecessary redundant parameters. Secondly, a new feature cycle selection method is used to establish a blast furnace stability index system. In this process, the importance score of each feature is obtained based on the prediction error change analysis of the machine learning model, and the features with low importance are deleted cyclically based on the designed judgment rules and the actual needs of the on-site experts, and finally the construction of the blast furnace stability index system is realized. Compared with the traditional artificial parameter selection method based on expert experience, the present invention does not rely on expert experience knowledge, can automatically select the core key parameters affecting the blast furnace condition based on the process mechanism and data-driven method, and automatically construct a blast furnace stability index system. Through experimental testing of actual production data, the blast furnace stability evaluation model established based on the index system obtained in the present invention has a good stability evaluation effect, can meet the monitoring needs of on-site experts on blast furnace conditions, and can help on-site operators better monitor and control the operating status of the blast furnace, thereby improving production efficiency and product quality, reducing production costs and energy consumption, and achieving continuous improvement and optimization of the blast furnace production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram of the principle of a blast furnace condition stability evaluation model provided by an embodiment of the present invention;

[0044] Figure 2 A flow chart of a method for constructing a blast furnace stability index system provided by an embodiment of the present invention;

[0045] Figure 3 A curve diagram showing changes in model prediction accuracy during a cyclic feature elimination process provided by an embodiment of the present invention;

[0046] Figure 4 This is a diagram of the prediction effect of the XGBoost model under the blast furnace index system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0048] As the first step in steel production, blast furnace ironmaking is a continuous and complex production process. Raw materials such as iron ore and coke are continuously injected into the blast furnace from the top of the blast furnace through the distribution system. A mixed gas containing high-temperature hot air, coal powder, natural gas and other fuels enters the blast furnace from the tuyere. Coke burns with oxygen in the high-temperature hot air to generate blast furnace gas containing carbon monoxide and hydrogen. During the rising process, the blast furnace gas will react with the descending iron ore and other furnace materials to produce reduction reaction, thereby removing oxygen from the iron ore to produce molten iron, which is discharged from the iron outlet of the blast furnace hearth. The unreduced impurities in the iron ore, the ash in the coke and coal combine with fluxes such as limestone to produce slag, which is discharged from the slag outlet of the blast furnace hearth. Figure 1 As shown, for the evaluation of blast furnace condition stability, this embodiment provides a method for constructing a blast furnace index system based on process mechanism and data-driven. First, the parameters in the evaluation index system are divided into condition parameters, operating parameters, process parameters and result parameters. According to the process mechanism, the required features are determined from the above four categories of parameter features, and the redundant features are eliminated. The remaining features are used as candidate features of the blast furnace stability index system. For the candidate features, a new parameter weight analysis method is first used to analyze the influence weight of each index on the blast furnace stability, and the index with a higher influence weight is selected to form a stability index system; then, a stability evaluation model is constructed based on this index system, and the model results are used as the evaluation value of the index system; finally, the above process is subjected to a cyclic feedback correction until a blast furnace stability evaluation model with higher performance and its corresponding index system are obtained. Through this index system, a blast furnace stability evaluation model can be effectively established, so that on-site operators can pay attention to the changing trend of key parameters and timely feedback the current blast furnace stability evaluation results. The flow chart of the method of this embodiment is shown as follows Figure 2 As shown, the details are as follows.

[0049] Step 1: Obtain an initial dataset for the blast furnace ironmaking process ,in, is the feature vector composed of all collected items of the i-th sample, is the molten iron output of the i-th sample, n is the number of samples, m is the sample dimension, Represents m-dimensional real number space.

[0050] In this embodiment, during the blast furnace ironmaking process, data collected based on the blast furnace's charge distribution system, gas system, cooling system, air supply system, injection system, slag system, and various sensors distributed on the blast furnace body are used to construct an initial data set D using the actual production process data of a steel plant's blast furnace from January 1, 2021 to November 27, 2023, where the number of samples is n=1380, and the sample dimension (number of collected items) is m=251, that is, there are 251 feature parameters.

[0051] Step 2: Perform data preprocessing on the initial data set D in step 1.

[0052] Step 2.1: Classify and preliminarily select parameters for all collected items in the initial data set D.

[0053] Step 2.1.1: According to the classification standards of blast furnace production data, all collection items are divided into four categories: condition parameters, operation parameters, process parameters, and result parameters.

[0054] Table 1 Collection item information of blast furnace ironmaking production process data

[0055] Collection item name Classification Sintered ore composition_MgO Conditional parameters White Horse Ball Group_CaO Conditional parameters Acidic sinter_S Conditional parameters … Blowing rate Operation parameters Oxygen enrichment Operation parameters Coke Batch Weight Operation parameters … Molten iron temperature Process parameters Standard wind speed Process parameters Hot air pressure Process parameters ... Coke ratio Result Parameters Coal injection ratio Result Parameters Hot metal output Result Parameters

[0056] In this embodiment, the number of samples in the initial data set is 1380, and the sample dimension is 251. The collection items of blast furnace process data include: sinter composition, white horse pellets, acidic sinter, metallurgical coke, hot air temperature, oxygen enrichment, coke batch weight, injection rate, iron tapping times, molten iron temperature, standard wind speed, hot air pressure, furnace coke ratio, small coke ratio, coal injection ratio, molten iron output, etc. The detailed information of its parameter classification is shown in Table 1.

[0057] Step 2.1.2: According to the process mechanism, the required features are determined from the above four categories of parameter features, and the redundant features are eliminated. The remaining features are used as candidate features of the blast furnace stability index system to obtain the data set. , n is the number of samples, is the sample dimension after removing redundant features.

[0058] In this example, most of the ore-related condition parameters are redundant features, all operating parameters and some important process parameters and result parameters are required features, and all the remaining features are used as candidate features of the blast furnace stability index system. The final data set is Number of samples , sample dimension .

[0059] Step 2.2: For the dataset obtained in step 2.1 Perform data filtering and use the moving average filtering method to filter the original data to eliminate mutations and spikes in the original data, make the data smooth, and obtain a data set The filtering formula is as follows:

[0060] ;

[0061] Among them, k represents the window size of the moving average filter, represents the moving average, represents the feature column vector composed of the jth feature, .

[0062] Step 2.3: For the dataset obtained in step 2.2 Use MinMax to normalize and get the data set , and the dataset As the current sample set; the normalization formula is as follows:

[0063] ;

[0064] in, represents the jth feature of the i-th sample, ; represents the minimum value of the jth feature, represents the maximum value of the jth feature, for Normalized value.

[0065] Step 3: For the current sample set, calculate the importance score of each feature in the current sample set.

[0066] Step 3.1: Split the current sample set into a training set , test set In this embodiment, 75% of the samples are randomly selected as the training set. , and use the remaining 25% of samples as the test set .

[0067] Step 3.2: Use the training set Train a machine learning-based prediction model , and calculate the model In the test set The root mean square error (RMSE) on In this example, the XGBoost model is used as the prediction model .

[0068] Step 3.3: For each feature column vector in the current sample set , calculate the error change of this feature on the XGBoost model .

[0069] Step 3.3.1: Generate h groups of Gaussian noise with different intensities ,in , and For the specified noise mean and variance, Gaussian noise with different means and the same variance is added to the test set in sequence Each characteristic column vector of In the new data set .

[0070] Step 3.3.2: Represents the test set The characteristic column vector of The new dataset generated after adding Gaussian noise is used to calculate the trained machine learning model in the dataset. The root mean square error (RMSE) on .

[0071] Step 3.3.3: Repeat steps 3.3.1 and 3.3.2 t times and take the average of the results to get the error change of the feature:

[0072] ;

[0073] in ; .

[0074] In this embodiment, when adding Gaussian noise of different strengths to each feature, a mean range from Start to , a list of noise means with a step size of 0.5, and the standard deviation of the noise is fixed to , each noise level will be calculated repeatedly t=15 times to evaluate the impact of different noise levels on the feature under the XGBoost model.

[0075] Step 3.4: Get the error change of all features , and normalized, recorded as ,in , and then use the normalized result as the importance score of each feature.

[0076] Step 4: Establish a stability index system for blast furnace process data.

[0077] Step 4.1: For all features of the current sample set, find the feature column vector with the lowest feature importance score , the feature Remove from the current sample set, where And get a new data set .

[0078] Step 4.2: The dataset obtained in step 4.1 As the current sample set, and split into training sets , test set In this embodiment, 75% of the samples are randomly selected as the training set. , and use the remaining 25% of samples as the test set .

[0079] Step 4.3: Use the training set Training the model , computational model In the test set The prediction effect (i.e., the coefficient of determination) ), and record the feature combination of the current sample set; The formula is as follows:

[0080] ;

[0081] in, represents the molten iron output of the i-th sample, represents the predicted value of molten iron output of the i-th sample, Represents the mean of molten iron production of all samples.

[0082] Step 4.4: Determine whether the stopping criteria are met. If the stopping criteria are met, the final blast furnace stability index system will be output and step 5 will be entered; otherwise, return to step 3, continue to calculate the feature importance score for the new current sample set and delete the features with low importance.

[0083] Step 4.4.1: Get the coefficient of determination of the model calculated in step 4.3 Value (i.e., the model removes features The prediction effect on the subsequent data set is recorded as ), and compare it with the removed features The decision coefficient of the previous model Value (denoted as ) for comparison. , then return to step 3; otherwise, go to step 4.4.2.

[0084] Step 4.4.2: Determine the features to be removed The coefficient of determination after Is it below the threshold of the coefficient of determination set by the on-site experts? In this implementation case, the threshold is set 0.7. , then return to step 3; otherwise, it indicates that the feature is removed. The subsequent feature combination can no longer meet the accuracy requirements of the field experts for the stability evaluation model, so the feature The previous feature combination is regarded as the final blast furnace stability index system, and the cycle ends and enters step 5.

[0085] In this embodiment, the index system obtained by the method includes 24 parameters in total, of which 17 parameters are parameters determined by prior knowledge and 7 parameters are parameters determined by the method of this embodiment. The obtained blast furnace stability index system parameter set is shown in Table 2. Figure 3 As shown in FIG. 1 , the curve diagram of the change of the model prediction accuracy during the process of cyclically eliminating features in this embodiment shows that as the features are cyclically eliminated, The value continues to increase until the characteristic parameter decreases to less than 24. The value is lower than the threshold value of 0.7 set by the on-site experts. At this time, the corresponding 24 parameter index system is output as the stability index system of blast furnace process data.

[0086] Step 5: Based on the stability index system of blast furnace process data obtained in step 4, a blast furnace stability evaluation model is constructed using appropriate machine learning modeling methods.

[0087] The method for constructing a blast furnace stability index system based on process mechanism and data-driven of this embodiment is applied to the blast furnace ironmaking production process data collected in real time. The blast furnace stability index system constructed based on the 24 parameters shown in Table 2 uses the data of a blast furnace from January 1, 2021 to December 31, 2022 as the training samples of the model, and the data from September 1, 2023 to November 27, 2023 as the test samples of the model.

[0088] Table 2 Blast furnace ironmaking production stability index system

[0089] Importance Importance score Parameter name unit Parameter classification Remark 1 0.2150 Ore batch weight ton Operation parameters Candidate features 2 0.1428 Furnace body 22.105m cooling wall temperature ℃ Process parameters Candidate features 3 0.1036 Hot air temperature ℃ Operation parameters Required Features 4 0.0934 Coke ratio % Result Parameters Required Features … … … … … … 22 0.0061 Hot air pressure kPa Process parameters Required Features 23 0.0045 Blowing energy kJ / s Process parameters Required Features 24 0.0043 Average particle size of coke entering the furnace % Conditional parameters Candidate features

[0090] like Figure 4 As shown in the figure, under this indicator system, the predicted value curve of the XGBoost model for molten iron production is very close to the true value curve, which shows that the model can effectively track the changing trend of molten iron production and make accurate predictions on the molten iron production in the actual production process.

[0091] The blast furnace index system based on process mechanism and data drive in this embodiment is used in different Model training, including XGBoost, neural network, decision tree and random forest. The number of features of the optimal stability index system output by each algorithm is recorded as , respectively using 、 RMSE is used as the evaluation index of the model. The experimental results of each model are shown in Table 3.

[0092] It can be seen from Table 3 that the method of this embodiment can meet the set threshold under different machine learning prediction model training.

[0093] Table 3 Comparison of experimental results of each model

[0094] XGBoost Neural Networks Ridge Regression Random Forest 24 20 22 24 Threshold 0.7 0.7 0.7 0.7 <![CDATA[ R 2 ]]> 0.7386 0.7756 0.7409 0.7142 RMSE 0.0325 0.0550 0.0605 0.0628

[0095] The above analysis shows that the blast furnace index system construction method proposed in the present invention can, in combination with the process mechanism, effectively find out the feature combination that is more important to the blast furnace production process and eliminate redundant features. It can provide high-quality data for various tasks in the blast furnace ironmaking production process, ensuring the efficient production and stable operation of blast furnace ironmaking.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A method for constructing a blast furnace stability index system based on process mechanism and data-driven, characterized in that: The steps include: Step 1: Obtain an initial dataset for the blast furnace ironmaking process ,in, is the feature vector composed of all collected items of the i-th sample, is the molten iron output of the i-th sample, n is the number of samples, m is the sample dimension, represents m-dimensional real number space; Step 2: Perform data preprocessing on the initial data set D in step 1. First, perform parameter classification and preliminary selection on all the collected items in the initial data set D to obtain the data set. Then, the data set Perform data filtering and normalization to obtain the data set , and the dataset As the current sample set; the specific process of parameter classification and preliminary selection is: Step 2.1.1: According to the classification standard of blast furnace production data, all collected items are divided into four categories: condition parameters, operation parameters, process parameters and result parameters; Step 2.1.2: According to the process mechanism, the required features are determined from the above four categories of parameter features, and the redundant features are eliminated. The remaining features are used as candidate features of the blast furnace stability index system to obtain the data set. , n is the number of samples, is the sample dimension after removing redundant features; Step 3: For the current sample set, calculate the importance score of each feature in the current sample set; the specific steps are: Step 3.1: Split the current sample set into a training set and test set ; Step 3.2: Use the training set Train a machine learning-based prediction model , and calculate the model In the test set The root mean square error on ; Step 3.3: For each feature column vector in the current sample set , calculate the feature in the model The error variation on , get the error change of all features ; Step 3.4: Normalize the error variation of all features and record it as ,in ;Then the normalized result is used as the importance score of each feature; Step 4: Based on the importance scores of the current sample set and its features, a blast furnace stability index system is constructed cyclically; Step 5: Based on the stability index system of blast furnace process data obtained in step 4, a blast furnace stability evaluation model is constructed using appropriate machine learning modeling methods.

2. The method for constructing a blast furnace stability index system based on process mechanism and data-driven according to claim 1, characterized in that: The data filtering process in step 2 uses a moving average filtering method to filter the data set. The original data in the filter is filtered to eliminate the mutation and peak phenomenon in the original data, making the data smooth and obtaining the data set. ; The filtering formula is as follows: ； Among them, k represents the window size of the moving average filter, represents the moving average, n is the number of samples, represents the feature column vector composed of the jth feature, .

3. The method for constructing a blast furnace stability index system based on process mechanism and data drive according to claim 2, characterized in that: The normalization process in step 2 uses MinMax to filter the data set after the data is processed Do normalization and get the data set ; The normalization formula is as follows: ； in, represents the jth feature of the i-th sample, ; represents the minimum value of the jth feature, represents the maximum value of the jth feature, for Normalized value.

4. The method for constructing a blast furnace stability index system based on process mechanism and data-driven according to claim 1, characterized in that: In step 3.3, the error variation range The calculation method is: Step 3.3.1: Generate h groups of Gaussian noise with different intensities ,in , and For the specified noise mean and variance; Gaussian noise with different means and the same variance is added to the test set in sequence Each characteristic column vector of In the new data set ; Step 3.3.2: Dataset Represents the test set The characteristic column vector of Add Gaussian noise to generate a new dataset and calculate the trained machine learning model on the dataset The root mean square error on ; Step 3.3.3: Repeat steps 3.3.1 and 3.2.2 t times and take the mean of the results to obtain the feature vector The error variation is: ； in, ; .

5. The method for constructing a blast furnace stability index system based on process mechanism and data drive according to claim 4, characterized in that: The step 4 specifically comprises the following steps: Step 4.1: For all features of the current sample set, find the feature column vector with the lowest feature importance score , the feature Remove from the current sample set, where And get a new data set ; Step 4.2: The dataset obtained in step 4.1 As the current sample set, and split into training sets and test set ; Step 4.3: Use the training set Training the model , computational model In the test set The prediction effect is the coefficient of determination , and record the feature combination of the current sample set; The calculation formula is as follows: ； in, represents the molten iron output of the i-th sample, represents the predicted value of molten iron output of the i-th sample, represents the mean of molten iron production of all samples; Step 4.4: Determine whether the stopping criteria are met. If the stopping criteria are met, the final blast furnace stability index system will be output and go to step 5; otherwise, return to step 3, continue to calculate the feature importance score for the new current sample set and delete the features with low importance.

6. The method for constructing a blast furnace stability index system based on process mechanism and data-driven according to claim 5, characterized in that: The specific method of step 4.4 is: Step 4.4.1: Get the coefficient of determination of the model calculated in step 4.3 Value, that is, the model removes features The prediction effect on the subsequent data set is recorded as , and remove the feature The decision coefficient of the previous model value For comparison; if , then return to step 3; otherwise, go to step 4.4.2; Step 4.4.2: Determine the features to be removed The coefficient of determination after Is it below the threshold of the coefficient of determination set by the on-site experts? ;like , then return to step 3; otherwise, it indicates that the feature is removed. The subsequent feature combination can no longer meet the accuracy requirements of the field experts for the stability evaluation model, so the feature The previous feature combination is regarded as the final blast furnace stability index system, and the cycle ends and enters step 5.

Images