Furnace core temperature prediction method and equipment based on multi-dynamic division and storage equipment

Abstract

The invention provides a furnace core temperature prediction method and device based on multi-dynamic division and a storage device, relates to the field of intelligent manufacturing, and aims to carry out furnace core temperature space distribution characteristic analysis and smelting unit operation system analysis in order to solve the problem of numerous smelting units influencing furnace core temperature heat exchange. The distribution of multivariate parameters among a plurality of smelting units is represented in detail, and the spatial-temporal characteristics influencing the temperature of the furnace core are qualitatively described. According to the method, different operation states of the blast furnace are subjected to subclass division in a manner of dividing the operation states of the blast furnace by t-SNE dimensionality reduction and spectral clustering, and an optimal prediction effect is sought in each subclass, so that the problem of low prediction precision caused by different fluctuation conditions of target variables in different operation states of the blast furnace is solved. A multi-dynamic division-based furnace core temperature prediction device and a storage device are used for realizing a multi-dynamic division-based furnace core temperature prediction method.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing, and in particular to a furnace core temperature prediction method, device and storage device based on multi-dynamic partitioning. Background Art

[0002] Blast furnace ironmaking is the core link of steel production, and its operating status is directly related to the economic benefits and environmental protection of steel enterprises. In the ironmaking process, the state of the furnace hearth is particularly critical. The blast furnace hearth is the main place for iron ore reduction and molten iron generation. Its internal structure is complex, mainly including dead material columns and active areas. The dead material column is a fixed area formed by the sintering of the charge at high temperature, and its air permeability and liquid permeability directly affect the activity of the furnace hearth. There is a close relationship between the furnace core temperature and the air permeability and liquid permeability of the dead material column. When the dead material column has good air permeability and liquid permeability, coal gas and blast furnace slag can pass smoothly, the chemical reaction inside the furnace hearth can be fully carried out, and the furnace core temperature is high, indicating that the furnace hearth is active; on the contrary, if the dead material column has poor air permeability and liquid permeability, the flow of coal gas and blast furnace slag is blocked, the chemical reaction inside the furnace hearth is inhibited, and the furnace core temperature is low, indicating that the furnace hearth is inactive.

[0003] Therefore, by monitoring the furnace core temperature, the air and liquid permeability of the dead material column can be indirectly evaluated, and then the activity of the furnace can be judged. However, there are many smelting units that affect the blast furnace core temperature. The traditional furnace core temperature monitoring method considers fewer influencing dimensions and often does not work well for scenarios with multivariate nonlinearity and time delay. Summary of the invention

[0004] In order to solve the problems of large fluctuations in the core temperature data, insufficient model stability and low accuracy caused by frequent fluctuations in process parameters of the core temperature prediction model under different furnace conditions, the present invention provides a core temperature prediction method, device and storage device based on multi-dynamic partitioning. Through the analysis of the blast furnace process mechanism, the distribution of multivariate parameters between multiple smelting units is characterized in detail, and the spatiotemporal characteristics affecting the core temperature are qualitatively described. The selected characteristic parameter data is divided into subclasses under different operating states by combining the t-distribution random neighborhood embedding dimension reduction and spectral clustering methods. And because the core temperature has a certain lag in the state characterization of the furnace cylinder activity, the present invention selects different prediction algorithms suitable for each subclass for different subclasses to obtain the core temperature prediction value for the next hour.

[0005] A furnace core temperature prediction method based on multi-dynamic partitioning mainly includes:

[0006] S1: Based on the blast furnace process mechanism, the entire smelting process and the heat exchange mechanism inside the furnace are analyzed, and multiple smelting units that affect the furnace core temperature are considered to determine the influencing factors of the furnace core temperature, which include: wind temperature balance, cooling balance, gas flow distribution, gas flow heat, and temperature measurement data;

[0007] S2: According to the spatial distribution characteristics of the factors affecting the furnace core temperature and the analysis of the smelting unit operation system, the process variables affecting the furnace core temperature are determined: cooling water temperature difference, cooling water flow, cold air pressure, cold air temperature, total oxygen enrichment, hot air pressure, hot air temperature, CO content, CO2 content, furnace top pressure, pressure difference, permeability index, gas generation, and adjacent thermocouple temperature measurement data;

[0008] S3: Based on the above process variables, the t-SNE technology combined with spectral clustering is used to classify the blast furnace operation status and obtain subclasses;

[0009] S4: Use the time series prediction algorithm to find the optimal algorithm in each subcategory and obtain the prediction result of the furnace core temperature in the next hour.

[0010] A storage device stores instructions and data for implementing a furnace core temperature prediction method based on multiple dynamic partitions.

[0011] A furnace core temperature prediction device based on multiple dynamic partitions comprises: a processor and a storage device; the processor loads and executes instructions and data in the storage device to implement a furnace core temperature prediction method based on multiple dynamic partitions.

[0012] The technical solution provided by the present invention has the following beneficial effects: the present invention analyzes the spatial distribution characteristics of the furnace core temperature and the operation system of the smelting units in order to solve the problem of numerous smelting units that affect the heat exchange of the furnace core temperature, characterizes the distribution of multivariate parameters between multiple smelting units in detail, and qualitatively describes the spatiotemporal characteristics that affect the furnace core temperature. By using t-SNE dimension reduction and spectral clustering to divide the blast furnace operating state, the different operating states of the blast furnace are subclassified, and the optimal prediction effect is sought in each subclass, which improves the problem of low prediction accuracy caused by different fluctuations of the target variable under different operating states of the blast furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0014] Figure 1 It is a flow chart of a furnace core temperature prediction method based on multiple dynamic partitions in an embodiment of the present invention.

[0015] Figure 2 It is an analysis diagram of the blast furnace process mechanism in an embodiment of the present invention.

[0016] Figure 3 It is a thermocouple distribution diagram of the 7.394m furnace core temperature detection point and surrounding temperature measurement points of No. 7 blast furnace in a certain factory in an embodiment of the present invention.

[0017] Figure 4Schematic diagram of factors affecting the furnace core temperature in an embodiment of the present invention.

[0018] Figure 5 It is a clustering effect diagram in an embodiment of the present invention.

[0019] Figure 6 It is a benchmark GRU effect diagram in an embodiment of the present invention.

[0020] Figure 7 It is a comparison chart of the prediction effects of three models under the operating status subclass in an embodiment of the present invention.

[0021] Figure 8 This is a comparison chart of the prediction effects of three models under the operating status subcategory 2 in an embodiment of the present invention.

[0022] Fig. 9 It is a comparison diagram of the prediction effects of three models under the operating state subcategory three in the embodiment of the present invention.

[0023] Fig.10 It is a schematic diagram of the operation of the hardware device in the embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0025] Aiming at the problems of the overall spatial interference of blast furnace and poor recognition effect in the current prediction of furnace core temperature under different furnace condition fluctuations, the present invention studies the spatial distribution characteristics of furnace core temperature and the operation system of smelting unit based on blast furnace process mechanism, characterizes the distribution of multivariate parameters among multiple smelting units in detail, qualitatively describes the spatiotemporal characteristics affecting the furnace core temperature, dynamically divides subclasses according to different blast furnace operation states, and adapts the optimal algorithm in the subclasses, and proposes a furnace core temperature prediction method based on multi-dynamic division, such as Figure 1 As shown, specifically including:

[0026] S1: Based on the blast furnace process mechanism, the entire smelting process and the heat exchange mechanism inside the furnace are analyzed, and multiple smelting units that affect the furnace core temperature are considered to determine the influencing factors of the furnace core temperature, which include: wind temperature balance, cooling balance, gas flow distribution, gas flow heat, and temperature measurement data;

[0027] S2: According to the spatial distribution characteristics of the factors affecting the core temperature and the analysis of the smelting unit operation system, the process variables affecting the core temperature are determined: cooling water temperature difference, cooling water flow, cold air pressure, cold air temperature, total oxygen enrichment, hot air pressure, hot air temperature, CO content, CO2 content, furnace top pressure, pressure difference, permeability index, gas generation, and adjacent thermocouple temperature measurement data, as shown in Table 1. The spatial distribution characteristics of the factors affecting the core temperature are that the blast furnace is a longitudinal device, and the top charging system, the bottom blowing system and air supply process, the cooling wall temperature of the entire furnace body, and the oxygen enrichment inside the furnace will affect the core temperature;

[0028] S3: Based on the above process variables, the t-SNE technology combined with spectral clustering is used to classify the blast furnace operation status and obtain subclasses;

[0029] S4: Use the time series prediction algorithm to find the optimal algorithm in each subcategory and obtain the prediction result of the furnace core temperature in the next hour.

[0030] Based on the blast furnace process mechanism, the present invention analyzes the entire smelting process and the heat exchange mechanism inside the furnace, and it can be found that the furnace core temperature is affected by multi-dimensional variables. Since the reaction time of the charge from top to bottom is difficult to determine, and the temporal and spatial distribution of different smelting units are different, it is necessary to comprehensively consider the multiple smelting units that affect the furnace core temperature. The specific process flow of blast furnace ironmaking is as follows: Figure 2 As shown. Coke, iron ore and solid flux are loaded from the top of the furnace. Auxiliary fuels such as oxygen-enriched air are blown into the blast furnace from the tuyere, and the rising airflow fully contacts the falling materials. The iron ore in the charge eventually produces liquid pig iron, and its heat change is significantly reflected in the furnace core temperature. Figure 3 This is the thermocouple distribution diagram of the 7.394m core temperature detection point and surrounding temperature detection points of the No. 7 blast furnace in a certain factory.

[0031] The prediction of blast furnace core temperature involves many parameters, so it is necessary to conduct qualitative research on these variables before analysis. Blast furnace production not only includes the main reaction process of the blast furnace, but also covers multiple units such as feeding, air supply, coal powder injection, recovery and slag iron treatment. When analyzing and selecting variables in the spatial dimension, we should identify those process parameters that are closely related to the furnace lining temperature although they are distributed in different smelting units based on process principles.

[0032] In the blast furnace smelting process, the heat balance can be divided into two types: dynamic and static. Figure 4 The influencing factors shown, material balance, wind temperature and cooling balance, and heat carried away by slag iron are relatively stable in the short term and can be regarded as static heat balance. However, the distribution of furnace charge and the change of coal gas flow are frequently changing, which belongs to dynamic heat balance. By monitoring the changing trend of these variables, the heat exchange analysis of furnace core temperature is carried out, and then the furnace core temperature is predicted.

[0033] Based on the heat balance of smelting and relevant expert experience, the present invention selects the state parameters and operating parameters shown in Table 1 as prediction variables for furnace core temperature prediction.

[0034] Table 1 Analysis of the mechanism of furnace core temperature affecting dimensions

[0035]

[0036] Blast furnace ironmaking involves the collaborative operation of many subsystems, and the state fluctuations of each subsystem may affect the performance of the entire blast furnace. Therefore, it is necessary to construct corresponding sub-models for the characteristics of each operation stage. In actual operation, the operating state of the blast furnace is not strictly distinguished. The present invention classifies the blast furnace operating state by using the process variables obtained by pretreatment and the method combining t-SNE technology with spectral clustering. These classified data are used as the input of the sub-model to construct a prediction model for the blast furnace core temperature.

[0037] (1) t-SNE Dimensionality Reduction

[0038] Based on the in-depth analysis and processing of blast furnace ironmaking process data, the collected blast furnace ironmaking process data can be summarized into a set Indicates that each vector contains M n-dimensional points, x1, x2, ..., x n They are cold air temperature, cold air pressure, total oxygen enrichment, etc. The t-SNE method works by calculating the KL divergence (Kullback-Leibler divergence) of probability values ​​based on the assumption that the data follows a Gaussian distribution in the high-dimensional space and a t-distribution in the low-dimensional space. Its goal is to minimize the KL divergence between the two distributions in the high-dimensional and low-dimensional spaces, which is calculated as follows:

[0039]

[0040] Both P and Q are n×n matrices that describe the probability distribution of point sets in high-dimensional and low-dimensional spaces, and their elements are p ij and q ij ; KL() represents KL divergence, F represents the objective function, i and j are dimensional data points, i = 1, 2, 3, ..., n, j = 1, 2, 3, ..., n. The calculation method is as follows:

[0041]

[0042] Among them, y is a point in the low-dimensional space, k and l represent any two different points in the low-dimensional space, and p ij is x in high-dimensional space i With x j The probability between ij is y in low-dimensional spacei With y j The probability between j|i Represents x j is x i The probability of the field, p i|j Represents x i is x j The probability of the field.

[0043] p j|i and p i|j The calculation method is:

[0044]

[0045]

[0046] Among them, x is a point in high-dimensional space, m is the number of nearest neighbors of each point, σ i , σ j Reflects the sample x i 、x j The surrounding sample distribution density, x m is the number of nearest neighbors of a point x.

[0047] The calculation formula of the stochastic gradient descent method is:

[0048]

[0049] The stochastic gradient descent method is used to iteratively optimize the intermediate variable objective function F. The low-dimensional data representation is used as the optimization target, and the position of the low-dimensional data points is continuously adjusted until the optimal solution is reached. The update rule is shown in formula (7):

[0050]

[0051] Where t is the number of iterations, η is the learning rate, α(t) is the learning momentum, and the iterative operation is completed when F reaches the preset accuracy. Y represents the output after iteration, Y t represents the output after the tth iteration, Y t-1 represents the output after the t-1th iteration, Y t-2 Represents the output after the t-2th iteration.

[0052] (2) Spectral clustering classification of blast furnace operation status

[0053] The present invention adopts a spectral clustering method based on a normalized Laplace matrix. After completing the data preprocessing and t-SNE dimension reduction steps, the present invention uses the spectral clustering method to subgroup the data. In order to determine the optimal number of clusters, the present invention uses the silhouette coefficient (SC) and the sum of squared errors (SSE) as evaluation indicators. The optimal number of clusters is determined by these indicator values, with the aim of maximizing the similarity of data within the same cluster and minimizing the similarity of data between different clusters. The evaluation functions of SC and SSE are defined as follows:

[0054]

[0055] Among them, a and b represent the average distances of sample i to all samples in the same cluster and different clusters. The value range of SC(i) is -1 to 1. The higher the index value, the better the clustering effect, the smaller the intra-class distance, and the larger the inter-class distance.

[0056]

[0057] Among them, C i represents the i-th sample cluster, p is C i The sample points, m i It is C i The centroid of , SSE is the clustering error of all samples, and k is the number of sample clusters, that is, the number of clusters. As the number of clusters k increases, the sample division will be more refined, the degree of aggregation of each cluster will gradually increase, and the error square sum SSE will gradually decrease.

[0058] Finally, the present invention uses TCN (Temporal Convolutional Network), GRU (Gated Recurrent Unit), LSTM (Long Short-Term Memory) algorithms to seek the optimal algorithm in each subclass and obtain the furnace core temperature prediction result. In order to accurately evaluate the prediction effect of the entire furnace core temperature prediction model of the present invention, the present invention uses two evaluation indicators, mean absolute error (MAE) and root mean square error (RMSE), for consideration. The specific calculation formula is as follows:

[0059]

[0060] Where n represents the number of sample data. represents the data mean, y i Indicates the true value of the data.

[0061] The present invention selects 20,000 actual blast furnace smelting parameter samples of No. 7 blast furnace of a certain factory from March 1, 2024 to June 1, 2024 as the original data set for this experiment for experimental research and functional verification. After data preprocessing, during the training process, the target iteration number is set to 100 times and the initial learning rate is set to 0.001.

[0062] In order to prove the advanced nature of the technology proposed in the present invention, the recognition effect is compared with the unimproved baseline model GRU, the GRU, LSTM, and TCN models under each subclass with multiple dynamic divisions.

[0063] First, after dimensionality reduction through t-SNE, the optimal number of clusters k=3 is calculated through SC and SSE. Figure 5 This is a clustering effect diagram. Dimension1, 2, and 3 are the three dimensions of the axis. Class0 in the lower left corner is subclass one of the experiment, Class1 in the upper right corner is subclass two of the experiment, and Class2 in the upper left corner is subclass three of the experiment.

[0064] The baseline model GRU prediction results are as follows Figure 6 As shown in the figure, the vertical axis is the target variable value, the horizontal axis is the sample value, the solid line Real Values ​​is the real value of the target variable, and the dotted line GRU Predicted Values ​​is the GRU predicted value of the benchmark model. The prediction results of the GRU, LSTM, and TCN models in the three subclasses (Class0, 1, and 2 correspond to subclasses one, two, and three respectively) after t-SNE dimension reduction and spectral clustering are shown in the figure Figure 7 , 8 , 9, the solid line Real Values ​​in each figure is the real value of the target variable, and the dotted lines GRU Predicted Values, LSTM Predicted Values, and TCN Predicted Values ​​are the predicted values ​​of GRU, LSTM, and TCN. The MAE and RMSE calculation results of the above methods are shown in Table 2.

[0065] Table 2 Comparison of training model performance

[0066]

[0067] Combination Figure 6 , 7 , 8, 9 and Table 2 show that the LSTM algorithm has the best prediction effect in subclass 1, the GRU has the best prediction effect in subclass 2, and the TCN has the best prediction effect in subclass 3. The comprehensive MAE and comprehensive RMSE of the optimal algorithms of the three subclasses are 0.110 and 0.144, respectively, which are better than the benchmark model GRU. The MAE and RMSE of the optimal algorithms of each subclass are better than those of the benchmark model GRU.

[0068] This is because the blast furnace will adjust different operating states according to economic benefits. The ironmaking process is completed by the coordination of many subsystems. Therefore, any change in the operating state of any subsystem will eventually affect the operating state of the blast furnace. In actual production, there is no clear dividing line for the division of the operating state of the blast furnace. Prediction without dynamic division will cause the model to learn too many unnecessary features, reducing its generalization ability. Dynamic division through t-SNE dimensionality reduction and spectral clustering, and optimization of prediction algorithms in each subclass can make the model more suitable for the subclass operating state. Through verification in the test set, the results show that the prediction effect has been significantly improved.

[0069] In summary, the method proposed in the patent effectively solves the problems of difficult identification, misidentification and low identification accuracy caused by complex interference of the furnace core temperature heat exchange smelting unit and different fluctuations in different operating conditions of the blast furnace, and has been verified in engineering practice.

[0070] See also Fig.10 , Fig.10 4 is a schematic diagram of the working of the hardware device of an embodiment of the present invention, wherein the hardware device specifically comprises: a furnace core temperature prediction device 401 based on multiple dynamic partitions, a processor 402 and a storage device 403.

[0071] A furnace core temperature prediction device 401 based on multiple dynamic partitions: The furnace core temperature prediction device 401 based on multiple dynamic partitions implements the furnace core temperature prediction method based on multiple dynamic partitions.

[0072] Processor 402: The processor 402 loads and executes the instructions and data in the storage device 403 to implement the furnace core temperature prediction method based on multiple dynamic partitions.

[0073] Storage device 403: The storage device 403 stores instructions and data; the storage device 403 is used to implement the furnace core temperature prediction method based on multiple dynamic partitions.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A furnace core temperature prediction method based on multi-dynamic partitioning, characterized in that: include: S1: Based on the blast furnace process mechanism, the entire smelting process and the heat exchange mechanism inside the furnace are analyzed, and multiple smelting units that affect the furnace core temperature are considered to determine the influencing factors of the furnace core temperature, which include: wind temperature balance, cooling balance, gas flow distribution, gas flow heat, and temperature measurement data; S2: According to the spatial distribution characteristics of the factors affecting the furnace core temperature and the analysis of the smelting unit operation system, the process variables affecting the furnace core temperature are determined: cooling water temperature difference, cooling water flow, cold air pressure, cold air temperature, total oxygen enrichment, hot air pressure, hot air temperature, CO content, CO2 content, furnace top pressure, pressure difference, permeability index, gas generation, and adjacent thermocouple temperature measurement data; S3: Based on the above process variables, the t-SNE technology combined with spectral clustering is used to classify the blast furnace operation status and obtain subclasses; S4: Use the time series prediction algorithm to find the optimal algorithm in each subcategory and obtain the prediction result of the furnace core temperature in the next hour.

2. A furnace core temperature prediction method based on multiple dynamic partitions as claimed in claim 1, characterized in that: Step S3 is specifically as follows: (1) t-SNE Dimensionality Reduction The t-SNE method is used to minimize the KL divergence between two distributions in high-dimensional and low-dimensional space, which is calculated as follows: Among them, P and Q are both n×n matrices, which are used to describe the probability distribution of point sets in high-dimensional and low-dimensional spaces, and their elements are p and ij and q ij ; KL() represents KL divergence, F represents the objective function, i and j are dimensional data points, i = 1, 2, 3, ..., n, j = 1, 2, 3, ..., n; Through the stochastic gradient descent method, the intermediate variable objective function F is iteratively optimized, the low-dimensional data representation is used as the optimization target, and the position of the low-dimensional data points is continuously adjusted until the optimal solution is reached; (2) Spectral clustering classification of blast furnace operation status The spectral clustering method is used to divide the data into subgroups, and the evaluation functions of SC and SSE are used to determine the optimal number of clusters. The evaluation functions of SC and SSE are defined as follows: Among them, a(i) and b(i) represent the average distances from sample i to all samples in the same cluster and different clusters. The higher the SC(i) value, the better the clustering effect, the smaller the intra-class distance, and the larger the inter-class distance. Among them, C i represents the i-th sample cluster, p is C i The sample points, m i It is C i The centroid of , SSE is the clustering error of all samples, and k is the number of sample clusters.

3. A furnace core temperature prediction method based on multiple dynamic partitions as claimed in claim 2, characterized in that: p ij and q ij The calculation method is as follows: Among them, x is a point in the high-dimensional space, y is a point in the low-dimensional space, k and l represent any two different points in the low-dimensional space, and p ij is x in high-dimensional space i With x j The probability between ij is y in low-dimensional space i With y j The probability between j|i Represents x j is x i The probability of the field, p i|j Represents x i is x j The probability of the field.

4. A furnace core temperature prediction method based on multiple dynamic partitions as claimed in claim 3, characterized in that: p j|i and p i|j The calculation method is: Among them, x is a point in high-dimensional space, m is the number of nearest neighbors of each point, σ i , σ j Reflects the sample x i 、x j The surrounding sample distribution density, x m is the number of nearest neighbors of a point x.

5. The method for predicting furnace core temperature based on multiple dynamic partitions according to claim 2, characterized in that: The update rule for iterative optimization is as follows: Where t is the number of iterations, η is the learning rate, α(t) is the learning momentum, and the iterative optimization is completed when the final objective function F reaches the preset accuracy. Y represents the output after iteration, Y t represents the output after the tth iteration, Y t-1 represents the output after the t-1th iteration, Y t-2 Represents the output after the t-2th iteration.

6. The method for predicting furnace core temperature based on multiple dynamic partitions according to claim 1, characterized in that: In order to evaluate the prediction effect, two evaluation indicators, MAE and RMSE, are used for consideration. The specific calculation formula is as follows: Where n represents the number of sample data. represents the data mean, y i Indicates the true value of the data.

7. A storage device, characterized in that: The storage device stores instructions and data for implementing the furnace core temperature prediction method based on multiple dynamic partitions as described in any one of claims 1 to 6.

8. A furnace core temperature prediction device based on multi-dynamic partitioning, characterized in that: include: A processor and a storage device; the processor loads and executes instructions and data in the storage device to implement the furnace core temperature prediction method based on multiple dynamic partitions as described in any one of claims 1 to 6.

Images