A blast furnace operation furnace type self-judgment method based on big data

Abstract

The application discloses a blast furnace operation furnace type self-judgment method based on big data, which comprises the following steps: taking the historical cooling wall thermocouple temperature of a blast furnace as a data set, and performing outlier rejection and missing value filling on the data set; performing clustering analysis on the processed data according to K-Means and TwoStep clustering algorithms, and obtaining operation furnace type classification results of different clustering cluster numbers; selecting Davies-Bouldin index and Dunn index evaluation indexes, and evaluating the operation furnace type classification results of different clustering cluster numbers; according to the evaluation results, determining the optimal clustering algorithm and the best clustering cluster number, and obtaining the optimal operation furnace type clustering result; and the new cooling wall thermocouple temperature is dispatched or a new class is established by calculating the distance between the new cooling wall thermocouple temperature and each clustering center, and the blast furnace operation furnace type is updated according to the data dispatching result. The application realizes the selection and dynamic updating of the optimal classification result of the blast furnace operation furnace type, provides guarantee for the production monitoring of the blast furnace operation furnace type, and is beneficial to high-quality, low-consumption, high-yield and smooth operation of the blast furnace.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing in the iron and steel metallurgy industry, and in particular to a method for self-determination of blast furnace operation type based on big data. Background Technology

[0002] The blast furnace operating configuration is the internal shape of the blast furnace that gradually evolves from the designed configuration after the blast furnace is put into production, through lining erosion and slag formation. During blast furnace smelting, operators mostly rely on cooling wall parameters, operating parameters, and production experience to indirectly analyze changes in the blast furnace operating configuration and thus judge the furnace condition. To ensure high-quality, low-consumption, high-yield, and long-life blast furnace production, a reasonable blast furnace operating configuration is necessary.

[0003] To address this, scholars both domestically and internationally have utilized big data analytics to conduct research on characterization methods for blast furnace operation types, focusing on the temperature of thermocouples on the cooling walls. Among these methods, cluster analysis is a commonly used approach, demonstrating good application results in actual blast furnace production. Blast furnace operators can use cluster analysis of the thermocouple temperatures on the cooling walls to obtain the temperature distribution of different cooling wall sections within different categories, thereby characterizing the blast furnace operation type based on the thermocouple temperatures. However, the clustering characterization methods for blast furnace operation types are still limited by the complexities of blast furnace smelting production and the understanding of clustering methods. The universality of different blast furnaces and fluctuations in blast furnace conditions restrict the application of blast furnace operation types, still relying on the experience of blast furnace operators to adjust the specific conditions of the blast furnace operation type. Therefore, the clustering characterization methods for blast furnace operation types are not entirely satisfactory.

[0004] The shortcomings of existing characterization methods are as follows: 1. CN201610621997.1, a classification method for blast furnace operation types in the blast furnace ironmaking process, obtains the blast furnace operation type corresponding to each type of temperature data by clustering the comprehensive coke ratio, qualified pig iron quantity, and ventilation volume. However, this method uses the traditional Laida criterion for outlier data, judging whether the data is outlier from a global perspective. Normal data during blast furnace condition fluctuations may be misjudged. In addition, this method does not clearly determine the method for cluster number and still relies on production experience, which has great limitations for blast furnace applications. 2. CN201510916529.2, a method for establishing a blast furnace operation model, uses historical smelting data as a dataset for cluster analysis and proposes an idea for establishing a blast furnace operation model. Although the static blast furnace operation model effectively characterizes the historical smelting state of the blast furnace, it does not consider the dynamic changes of the operation model, and the information value in the blast furnace smelting data is not fully explored, resulting in a certain gap with actual production. 3. In the study of blast furnace operation types, K-Means and TwoStep clustering algorithms are mainly used. However, the application effects of different clustering algorithms are not clear, and how to determine the optimal clustering algorithm still needs to be solved. Summary of the Invention

[0005] The objective of this invention is achieved through the following technical solutions.

[0006] Based on the above shortcomings, the purpose of this invention is to provide a blast furnace operation type self-determination method based on big data. This method uses big data analysis to clean and fill blast furnace smelting data, automatically determine the optimal classification of blast furnace operation types, and automatically update the original classification of operation types based on new data on the basis of static operation types.

[0007] The technical solution provided by this invention is as follows:

[0008] A blast furnace operation type self-determination method based on big data includes the following steps:

[0009] Step 1: Using the historical thermocouple temperatures of the blast furnace cooling wall as the dataset, filter out outliers in the dataset according to the improved Laida criterion and delete outlier data.

[0010] Step 2: For data missing due to thermocouple failure and outliers that have been deleted, perform data imputation on the dataset by using linear interpolation.

[0011] Step 3: Perform cluster analysis on the processed dataset using K-Means and TwoStep clustering algorithms to obtain the classification results of different numbers of clusters for different furnace types;

[0012] Step 4: Use the Davies-Bouldin index and Dunn index as evaluation metrics to evaluate the classification results of different cluster numbers for different furnace types;

[0013] Step 5: Based on the evaluation results, determine the optimal clustering algorithm and the optimal number of clusters to obtain the optimal operating furnace type clustering result;

[0014] Step 6: Data is assigned or new classes are established by calculating the new cooling wall thermocouple temperature and the distance to each cluster center, and the blast furnace operation type is updated according to the data assignment results.

[0015] Furthermore, the improved Laida criterion in step 1 adds an observation window of 10 minutes in length to the observation values ​​selected by the global Laida criterion. Based on the comparison results between the observation values ​​and the data within the observation window, it is determined whether the observation value is normal fluctuation data or abnormal data of the blast furnace.

[0016] Furthermore, in step 3, K-Means and TwoStep clustering algorithms are selected based on whether the clustering algorithm is suitable for large datasets, time complexity, and space complexity.

[0017] Furthermore, the number of clusters is limited to 5 to 12.

[0018] Furthermore, the Davies-Bouldin index and Dunn index evaluation metrics in step 4 can characterize the intra-class and inter-class effects of the clustering results.

[0019] Furthermore, in step 5, based on the evaluation results, the adaptability of different clustering algorithms to the dataset and the optimal number of clusters are determined, thereby determining the optimal operating furnace type clustering result.

[0020] Furthermore, the smaller the Davies-Bouldin index and the larger the Dunn index, the greater the differences between classes and the more similar the data within classes, indicating better clustering performance.

[0021] Further, step 6 specifically includes: calculating the distance between new data points in the new dataset and each cluster center, and setting a threshold for each class; assigning new data points to the class with the smallest distance; if the distance between a new data point and a cluster center is less than the threshold, the new data point belongs to that class; if the distance between a new data point and a cluster center is greater than the threshold, the new data point does not belong to that class, and a new class is created.

[0022] Furthermore, the threshold for each category is the farthest distance between all data in the original category and the class center.

[0023] Furthermore, after the data distribution is completed, the cluster center is reset according to the proportion of the new data in the original data, thereby updating the blast furnace operation furnace type.

[0024] The advantages of this invention are:

[0025] This invention provides a self-determination method for blast furnace operation types based on big data. This method achieves the selection of the optimal classification result for blast furnace operation types and the dynamic updating of blast furnace operation types, providing assurance for production monitoring of blast furnace operation types and contributing to high-quality, low-consumption, high-yield, and smooth operation of blast furnaces. The beneficial effects of this invention are:

[0026] 1. The outlier screening method has been optimized, and the improved Raida criterion is more adaptable to data fluctuations when blast furnace conditions fluctuate;

[0027] 2. By using the Davies-Bouldin index and Dunn index clustering evaluation metrics, the application effects of different clustering algorithms and the clustering results with different numbers of clusters were compared to obtain the optimal clustering result for the operating furnace type, thus achieving accurate classification of blast furnace operating furnace types;

[0028] 3. For new data, data is assigned or new classes are established based on its distance from the original cluster center, and the blast furnace operation type is updated based on the data assignment results, thus realizing the dynamic classification of blast furnace operation types. Attached Figure Description

[0029] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0030] Figure 1 This is a flowchart illustrating a self-determination method for blast furnace operation based on big data.

[0031] Figure 2 A schematic diagram of the decision algorithm for classifying, optimizing, and updating blast furnace operation furnace types;

[0032] Figure 3 This is a schematic diagram showing the distribution of thermocouple locations on the blast furnace cooling wall in the dataset used in Example 1;

[0033] Figure 4 This is a comparison chart showing the optimization effect of the method for filtering local outliers in Example 1;

[0034] Figure 5 This is the linear interpolation fill-in result for local data loss in Example 1;

[0035] Figure 6 The evaluation results of the Davies-Bouldin index and Dunn index of the clustering results in Example 1;

[0036] Figure 7 This is the clustering result of the optimal operating furnace type in Example 1. Detailed Implementation

[0037] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0038] Example 1 provides a flowchart of a blast furnace operation type self-determination method based on big data, such as... Figure 1 As shown in the diagram, the flowchart of the decision algorithm for classifying, selecting, and updating blast furnace operation furnace types is as follows: Figure 2 As shown, the specific steps are as follows:

[0039] Step 1: Using the historical thermocouple temperatures of the blast furnace cooling walls as the dataset, outlier data is filtered and deleted according to the improved Laida criterion.

[0040] In this embodiment, the dataset consists of the historical thermocouple temperatures of the blast furnace cooling walls from June 1, 2015 to July 28, 2015, at a steel plant in China. The thermocouples are distributed in sections 6 to 11 of the cooling walls. Figure 3 As shown.

[0041] In this embodiment, an improved Laida criterion is used to filter out outliers in the dataset. The traditional Laida criterion calculation method is as follows:

[0042] |V i |=|x i -x|>3σ

[0043] In the formula: x is the arithmetic mean of all data, V i σ represents the residual error, and σ represents the standard deviation.

[0044] If a certain measurement data V i If (1≤i≤n) satisfies the above formula, then the measured value x is considered to be... iThese should be removed. This invention optimizes the traditional Laida criterion by adding a 10-minute observation window to the selected observations. Based on the comparison between the observed value and the average value within the observation window, it determines whether the observed value is due to normal fluctuations in the blast furnace or is abnormal data. The optimization effect of the method for filtering local data outliers is compared to... Figure 4 As shown.

[0045] Step 2: For data missing due to thermocouple failure and outliers that have been deleted, perform data imputation on the dataset by using linear interpolation.

[0046] In this example, missing values ​​(including data missing due to thermocouple failure and deleted outliers) are filled using linear interpolation. Taking the local data selected in Step 1 as an example, the result after data filling is as follows: Figure 5 As shown.

[0047] Step 3: Perform cluster analysis on the processed data using K-Means and TwoStep clustering algorithms to obtain the classification results of the furnace type with different numbers of clusters;

[0048] In this example, K-Means and TwoStep clustering algorithms are used to perform cluster analysis on the processed data to obtain the clustering results of blast furnace operation types when the number of clusters is 5 to 12.

[0049] Step 4: Use the Davies-Bouldin index and Dunn index as evaluation metrics to evaluate the classification results of different cluster numbers for different furnace types.

[0050] In this example, the Davies-Bouldin index and Dunn index are used to evaluate the classification results of operating furnace types with different numbers of clusters. The calculation methods for the Davies-Bouldin index (DBI) and Dunn index (DI) are as follows:

[0051]

[0052]

[0053] In the formula: k is the number of clusters, K is the total number of clusters, and w i ,w j As the cluster center, ||w i -w j ||2 represents the distance between the centers of each cluster. The average distance between all data within the same cluster, where m and n represent different clusters, and x... i ,xj For different data points within a cluster, ||x i -x j || represents the distance between any two data points. Ω m Ω n This represents the set of all data points within the clusters m and n.

[0054] Both the Davies-Bouldin indicator and the Dunn indicator are evaluation metrics for measuring inter-cluster dissimilarity and intra-cluster similarity. A smaller Davies-Bouldin indicator value indicates better clustering results, more significant inter-cluster differences, and more compact intra-cluster data, while the Dunn indicator shows the opposite. The clustering results for blast furnace operation types with 5 to 12 clusters were evaluated, and the evaluation results are as follows: Figure 6 As shown.

[0055] Figure 6 As can be seen, for the blast furnace cooling wall temperature dataset used, in the K-Means dataset with 5-12 clusters, the Davies-Bouldin indicator results are consistently lower than TwoStep's, while the Dunn indicator results are consistently higher than TwoStep's. This indicates that K-Means' clustering results are superior to TwoStep's. Furthermore, in the K-Means results with 8 clusters, the Davies-Bouldin indicator results are smaller, while the Dunn indicator results are larger, indicating better clustering performance.

[0056] Step 5: Based on the evaluation results, determine the optimal clustering algorithm and the optimal number of clusters to obtain the optimal operating furnace type clustering results;

[0057] In this example, the clustering results are analyzed based on the results of the clustering evaluation indicators. From Figure 6 As can be seen, for the blast furnace cooling wall temperature dataset used, within the cluster size range of 5-12, the Davies-Bouldin indicator results are consistently lower than TwoStep's, while the Dunn indicator results are consistently higher than TwoStep's. This indicates that K-Means' clustering results are superior to TwoStep's. Furthermore, when the number of clusters in the K-Means results is 8, the Davies-Bouldin indicator results are smaller, while the Dunn indicator results are larger, indicating better clustering performance. Therefore, for the dataset used in this example, the K-Means clustering algorithm performs better, with an optimal number of clusters of 8. The clustering results are as follows: Figure 7 As shown.

[0058] Step 6: The new cooling wall thermocouple temperatures are assigned to data or new classes are established by calculating their distance from each cluster center, and the blast furnace operation type is updated based on the data assignment results.

[0059] In this example, the cooling wall thermocouple temperatures from a domestic steel plant from June 1, 2015 to July 28, 2015 are used as the original dataset, and the data from July 29 to July 31 are imported as the new dataset to update the furnace type.

[0060] In this example, the distance between new data points in the new dataset and each cluster center is calculated, and a threshold is set for each class. The new data points are assigned to the class with the smallest distance. If the distance between a new data point and a cluster center is less than the threshold, the new data point belongs to that class. If the distance between a new data point and a cluster center is greater than the threshold, the new data point does not belong to that class, and a new class is created.

[0061] In this example, based on the distribution results of all data, the cluster centers are reset according to the proportion of the new data in the original cluster data to realize the update of the blast furnace operation furnace type. The cluster centers after the data update from July 29, 2015 to July 31, 2015 are shown in Table 1, while the results of re-clustering all data from June 1, 2015 to July 31, 2015 are shown in Table 2.

[0062] Table 1. Cluster centers after data update from July 29, 2015 to July 31, 2015

[0063]

[0064] Table 2. Results of re-clustering all data from June 1, 2015 to July 31, 2015

[0065]

[0066] In this example, as can be seen from Tables 1 and 2, the cluster centers obtained by assigning and updating new data based on the original blast furnace type classification results are similar to the results of re-clustering all data. This effectively represents the blast furnace type while ensuring good time and space complexity of the algorithm.

[0067] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A blast furnace operation furnace type self-determination method based on big data, characterized in that, Includes the following steps: Step 1: Using the historical thermocouple temperatures of the blast furnace cooling wall as the dataset, outlier data is filtered and deleted according to the improved Laida criterion. The improved Laida criterion in Step 1 adds an observation window of 10 minutes in length to the observation values ​​selected by the global Laida criterion. Based on the comparison between the observation values ​​and the data within the observation window, it is determined whether the observation value is normal fluctuation data of the blast furnace or abnormal data. Step 2: For data missing due to thermocouple failure and outliers that have been deleted, perform data imputation on the dataset by using linear interpolation. Step 3, according to K -Means and TwoStep clustering algorithms are used to perform cluster analysis on the processed dataset to obtain the furnace type classification results with different numbers of clusters; Step 4: Use the Davies-Bouldin index and Dunn index as evaluation metrics to evaluate the classification results of different cluster numbers for different furnace types; Step 5: Based on the evaluation results, determine the optimal clustering algorithm and the optimal number of clusters to obtain the optimal operating furnace type clustering result; Step 6: Assign data or establish new classes by calculating the distance between the new cooling wall thermocouple temperature and each cluster center, and update the blast furnace operation type according to the data assignment results; Step 6 specifically includes: calculating the distance between the new data point in the new dataset and each cluster center, and setting the threshold for each class; assigning the new data point to the class with the smallest distance. If the distance between the new data point and the cluster center is less than the threshold, the new data point belongs to that class; if the distance between the new data point and the cluster center is greater than the threshold, the new data point does not belong to that class, and a new class is established. The threshold for each category is the farthest distance between all data in the original category and the class center; After the data distribution is completed, the cluster center is reset according to the proportion of the new data in the original data to realize the update of the blast furnace operation furnace type; In step 3, the clustering algorithm is selected based on its suitability for large datasets, time complexity, and space complexity. K -Means and TwoStep clustering algorithms; The number of clusters is limited to 5 to 12. In step 4, the Davies-Bouldin index and Dunn index evaluation metrics can characterize the intra-class and inter-class effects of the clustering results. In step 5, based on the evaluation results, the adaptability of different clustering algorithms to the dataset and the optimal number of clusters are determined, thereby determining the optimal operating furnace type clustering result; The smaller the Davies-Bouldin index and the larger the Dunn index, the greater the differences between classes and the more similar the data within classes, indicating a better clustering effect.

Images