A method for predicting carbon emission concentration in blast furnaces based on digital twins

Abstract

This invention proposes a method for predicting carbon emission concentrations from blast furnaces based on digital twins, belonging to the field of digital twin technology. This invention employs multi-source data fusion and uses a simple mechanistic model to construct the blast furnace temperature field. By combining operating parameters with key features of the thermal imaging image of the blast furnace top and supplementing with temperature field features, a shared matrix is ​​established. Then, ResNet50 and an autoencoder are used to extract image features and establish a random forest (RF) model. Based on the shared matrix, a random weighted neural network algorithm (RVFLNs) model is constructed, and the prediction results of the random forest (RF) model and the random weighted neural network algorithm (RVFLNs) model are combined using a particle swarm optimization (PSO) algorithm, thereby improving prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically a method for predicting carbon emission concentrations from blast furnaces based on digital twins. Background Technology

[0002] Blast furnaces are among the most critical pieces of equipment in steel production. Their operational efficiency and emission control are essential for increasing steel output, saving energy, reducing emissions, and lowering production costs. With the introduction of national dual-carbon targets, accurately predicting the CO2 and CO emission concentrations of blast furnaces is crucial for the sustainable development of the steel industry and for achieving carbon neutrality. Researchers have proposed various methods for predicting blast furnace carbon emissions, which can be broadly categorized into two types: mechanistic models and data-driven models. The former relies on understanding and simulating the physical and chemical processes inside the blast furnace, while the latter utilizes statistical and machine learning techniques to analyze and predict emission patterns based on historical data.

[0003] Mechanistic modeling methods involve a deep understanding and simulation of complex physical and chemical processes within blast furnaces, enabling effective prediction of CO2 and CO concentrations in blast furnace emissions. Wu et al. used a variable activated energy model (VAEM) to conduct experimental research on the molten reduction reaction of coal coke and iron ore under simulated blast furnace conditions. By analyzing the kinetic behavior of the molten reduction process, they predicted carbon oxides in the blast furnace exhaust gas. Xia et al. established a mathematical model including the blast furnace shaft, combustion zone, and gas circulation system by analyzing heat transfer, combustion reactions, and gas circulation within the blast furnace to estimate the composition of blast furnace emissions. Li et al.6 adopted an all-oxygen blast furnace technology and developed a carbon metabolism and prediction model to study and analyze the changes in CO2 and CO in blast furnace emissions and their impact on carbon emissions. Jiang et al.7 constructed a mathematical model of the mass and energy balance of the blast furnace and a heat balance model of the hot blast stove. They used metallurgical thermodynamics methods to predict and analyze the composition of blast furnace gas.

[0004] While mechanistic modeling can effectively explain the carbon oxide production process, it also has some drawbacks. First, mechanistic models are poorly adaptable to changes in blast furnace operating conditions and are difficult to update in real time. Second, these models require significant computational resources, making them unsuitable for large-scale industrial applications.

[0005] In recent years, there has been an increasing number of studies using advanced data-driven models to predict blast furnace gas composition. These methods, through in-depth analysis of historical data and operating parameters, can provide more accurate and real-time predictions, thereby optimizing blast furnace operation. Zhang et al. proposed a temporal dual-graph convolutional network (TDGCN) model combining graph convolution, hypergraph convolution, and TimesNet to predict CO2 and CO content in blast furnace gas. Xiao et al. developed a chaotic radial basis function (RBF) model for predicting carbon monoxide utilization rate (CMUR). By reconstructing the phase space of CMUR, they achieved high-precision prediction using chaos theory. Sha et al. introduced a prediction method based on kernel principal component analysis (KPCA) and extreme learning machine (ELM) for carbon monoxide utilization rate. They selected important input variables using KPCA and developed a prediction model using ELM. Feng et al. proposed a data-driven model for predicting blast furnace gas production based on spectral decomposition, which significantly improved prediction accuracy through spectral decomposition and combined modeling. Yang et al.12 adopted a multi-factor prediction model based on generalized phase arrangement entropy (GPPE) and singular spectral decomposition (SSD), and combined it with a bidirectional long short-term memory network (GPPE-SSD-TDO-BiLSTM) optimized by an artificial Tasmanian devil optimizer to predict the CO2 and CO content in blast furnace gas.

[0006] While these methods have made significant progress in prediction accuracy and real-time performance, they have failed to fully utilize the collaborative modeling capabilities of top-view thermal imaging of blast furnaces. Top-view thermal imaging not only provides detailed information on the temperature distribution inside the blast furnace but also indirectly reflects combustion and emission patterns in different regions during operation. Ignoring this critical data source can lead to deficiencies in predictive models when dealing with carbon emission fluctuations and optimizing control strategies. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to propose a method for predicting blast furnace carbon emission concentration based on digital twins, comprising:

[0008] Step 1: Real-time acquisition of production data during the blast furnace smelting process. The production data includes operating parameters, tuyere parameters, and multiple thermal imaging images of the top of the blast furnace. The tuyere parameters include tuyere temperature and tuyere velocity.

[0009] Step 2: Filter all thermal images of the top of the blast furnace to obtain thermal images that are not obscured by the charging chute;

[0010] Step 3: Based on the thermal imaging image and vent parameters that are not obstructed by the loading chute, obtain the temperature field characteristics, which include the temperature gradient, maximum temperature and average temperature.

[0011] Step 4: For each thermal image not obscured by the loading chute, feature extraction is performed using ResNet50 and an autoencoder to obtain image features for all thermal images not obscured by the loading chute. These features are then compressed into an image feature matrix X. Independent component analysis (ICA) is used to extract non-Gaussian independent source features S from the image feature matrix X. Inverse ICA transform is then used to process the non-Gaussian independent source features S to obtain the key image features X. e X, the key features of the image e Combined with the physical variable Z, the shared matrix X is obtained. c =[Z;X e ], where the physical variable Z includes the preprocessed operating parameters and temperature field characteristics, and the principal component analysis algorithm is used to extract the Gaussian features in the shared matrix;

[0012] Step 5: Obtain the final carbon monoxide and carbon dioxide concentrations based on the non-Gaussian independent source characteristic S and the Gaussian characteristic.

[0013] Optionally, step 2 specifically includes:

[0014] By using the gray-level co-occurrence matrix, feature extraction is performed on each thermal image of the top of the blast furnace to obtain the texture features of each thermal image of the top of the blast furnace. The texture features include contrast, energy, uniformity, and pixel difference in the thermal image of the top of the blast furnace. Based on the thermal image of the top of the blast furnace and the corresponding texture features, all thermal images of the top of the blast furnace are filtered to obtain thermal images that are not obscured by the charging chute.

[0015] Optionally, based on the thermal imaging images of the blast furnace top and corresponding texture features, all thermal imaging images of the blast furnace top are filtered to obtain thermal imaging images that are not obscured by the charging chute, including:

[0016] Based on the thermal imaging image of the blast furnace top, the production stage of the blast furnace is determined. When the blast furnace is in the uncharged stage, the range of texture feature values ​​that are not obscured by the charging chute is obtained and used as the target range. When the blast furnace is in the charging stage, the range of texture feature values ​​that are not obscured by the charging chute is obtained and used as the target range. A thermal imaging image of the blast furnace top with texture features within the target range is obtained to obtain a thermal imaging image that is not obscured by the charging chute.

[0017] Optionally, step 3 specifically includes:

[0018] The mass conservation equation, momentum conservation equation, and energy conservation equation are used to process the vent parameters to obtain the initial temperature field. For the flame area in the thermal imaging image that is not blocked by the loading chute, the gray value of the flame area is extracted, and the actual temperature value corresponding to the gray value is obtained. The actual temperature value is combined with the initial temperature field to obtain the final temperature field. The final temperature field is then subjected to feature extraction to obtain the temperature field features.

[0019] Optionally, step 5 specifically includes:

[0020] Step 5.1: Using the first prediction model based on the random forest regression algorithm, process the non-Gaussian independent source feature S in Step 4 to obtain the first concentration prediction values ​​of carbon monoxide and carbon dioxide.

[0021] The first prediction model is obtained by training multiple first training samples. The first training samples include first input samples and first output samples. The first input samples are historical non-Gaussian independent source features. The first output samples include the true values ​​of carbon monoxide concentration and carbon dioxide concentration corresponding to the historical non-Gaussian independent source features.

[0022] Step 5.2: The Gaussian features in Step 4 are processed by multiple different second prediction models based on the random weighted neural network algorithm to obtain multiple second prediction values. The second prediction values ​​include the second concentration prediction values ​​of carbon monoxide and carbon dioxide. According to the preset weights, the second concentration prediction values ​​of all carbon monoxide are weighted and averaged to obtain the third concentration prediction value of carbon monoxide. The second concentration prediction values ​​of all carbon dioxide are weighted and averaged to obtain the third concentration prediction value of carbon dioxide.

[0023] Step 5.3: Using the Particle Swarm Optimization (PSO) algorithm, determine the first and second weights. Based on the first and second weights, perform a weighted sum of the first and third predicted carbon monoxide concentrations to obtain the final carbon monoxide concentration. Similarly, based on the first and second weights, perform a weighted sum of the first and third predicted carbon dioxide concentrations to obtain the final carbon dioxide concentration.

[0024] Optionally, in step 5.3, multiple different second prediction models are obtained through the following steps:

[0025] Step A1: Obtain historical Gaussian features and the corresponding true concentration values, including the true concentration values ​​of carbon monoxide and carbon dioxide. Use the historical Gaussian features as the second input sample and the true concentration values ​​as the second output sample. The second input sample and the second output sample form the second training sample, thereby obtaining multiple second training samples.

[0026] Step A2: Using a random sampling strategy with replacement, Bagging sampling is performed on multiple second training samples to obtain multiple second training samples, forming a subset of training sets, and thus obtaining multiple subsets of training sets.

[0027] Step A3: For each sub-training set, construct an initial model of a random weighted neural network algorithm, wherein the weights and biases of the initial model are random;

[0028] Step A4: Train the initial model using a sub-training set. Specifically, input samples from the sub-training set are input into the initial model to obtain historical predicted values ​​corresponding to historical Gaussian features. These historical predicted values ​​include predicted values ​​for carbon monoxide concentration and carbon dioxide concentration. Based on the historical predicted values ​​and output samples, a regularized least squares algorithm is used to update the parameters in the initial model until the model converges, resulting in a second prediction model. This process leads to multiple different second prediction models corresponding to multiple sub-training sets.

[0029] Optionally, in step 5.3, the first weight and the second weight are determined using the Particle Swarm Optimization (PSO) algorithm, including:

[0030] Step B1: Obtain the predicted value and the actual value of carbon monoxide concentration, as well as the predicted value and the actual value of carbon dioxide concentration. The predicted value of carbon monoxide concentration includes a first concentration value and a second concentration value, and the predicted value of carbon dioxide concentration includes a third concentration value and a fourth concentration value.

[0031] Step B2: Initialize the velocity and position of all particles in the particle swarm. The position of each particle includes two randomly initialized weights. Initialize the individual optimal position and the global optimal position, and set the initial number of iterations.

[0032] Step B3: Take the initial velocity of the particle as the current velocity, the initial position of the particle as the current position, the initial individual optimal position of each particle as the current individual optimal position of that particle, the initial global optimal position as the current global optimal position, and the initial number of iterations as the current number of iterations;

[0033] Step B4: For each particle, according to the weight in the particle's current position, the first concentration value and the second concentration value are weighted and summed to obtain the weighted predicted value of carbon monoxide concentration; the third concentration value and the fourth concentration value are weighted and summed to obtain the weighted predicted value of carbon dioxide concentration.

[0034] Step B5: For each particle, calculate the mean square error between the weighted predicted value of carbon monoxide concentration and the actual value of carbon monoxide concentration to obtain the first error, and calculate the mean square error between the weighted predicted value of carbon dioxide concentration and the actual value of carbon dioxide concentration to obtain the second error. The first error and the second error constitute the current fitness of the particle. Among the current fitness of the particle and the fitness corresponding to the current individual optimal position of the particle, select the position with the smaller fitness as the new current individual optimal position of the particle.

[0035] Step B6: Among the fitness of the current individual best position after all particles are updated, and the fitness of the current global best position, select the position with the smaller fitness as the new current global best position;

[0036] Step B7: Update the current position and current velocity of each particle, specifically using the following formula:

[0037] v i,t+1 =ωv i,t +c1r1(p best,i -x i,t )+c2r2(g best -x i,t )

[0038] x i,t+1 =x i,t +v i,t+1

[0039] Where ω is the inertia weight, c1 and c2 are the acceleration constants, r1 and r2 are random numbers, and v i,t+1 Let v be the current velocity of the i-th particle at iteration number t+1. i,t Let p be the current velocity of the i-th particle at the current iteration number t. best,i Let g be the optimal position of the i-th particle in the current individual at the current iteration number t. best Let x be the current globally optimal position of the i-th particle at the current iteration number t. i,t Let x be the current position of the i-th particle at the current iteration number t. i,t+1 Let be the current position of the i-th particle at iteration number t+1;

[0040] Step B8: Increment the current iteration count by one. If the current iteration count is less than the preset count, or the fitness of the new current global optimal position is less than the preset threshold, use the two weights in the new current global optimal position as the first weight and the second weight. If the current iteration count is not less than the preset count, and the fitness of the new current global optimal position is not less than the preset threshold, return to step B4 until the current iteration count is less than the preset count, or the fitness of the new current global optimal position is less than the preset threshold.

[0041] Optionally, an independent component analysis algorithm is used to extract the non-Gaussian independent source features S from the image feature matrix X, specifically achieved through the following formula:

[0042] S = WX;

[0043] Where W is the unmixing matrix.

[0044] Optionally, the non-Gaussian independent source features S can be processed using the inverse ICA transform to obtain the key image features X. e Specifically, this is achieved through the following formula:

[0045] X e＝ X-AS = (I-AW)X;

[0046] Where A is the mixing coefficient matrix and I is the identity matrix.

[0047] The beneficial effects of adopting the above technical solution are as follows:

[0048] This invention proposes a method for predicting carbon emission concentrations from blast furnaces based on digital twins. The method employs multi-source data fusion and uses a simple mechanistic model to construct the blast furnace temperature field. By combining operating parameters with key features from a thermal image of the blast furnace top and supplementing this with temperature field features, a shared matrix is ​​established. Then, ResNet50 and an autoencoder are used to extract image features and establish a random forest (RF) model. Based on the shared matrix, a random weighted neural network algorithm (RVFLNs) model is constructed, and a particle swarm optimization (PSO) algorithm is used to combine the prediction results of the random forest (RF) model and the random weighted neural network algorithm (RVFLNs) model, thereby improving prediction accuracy. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a method for predicting carbon emission concentration in a blast furnace based on digital twins, as described in an embodiment of the present invention.

[0050] Figure 2These are thermal imaging images of the blast furnace without charging in an embodiment of the present invention. Among them, Figure (a) is the thermal imaging image of the top of the blast furnace in the first frame when it is not charged, Figure (b) is the thermal imaging image of the top of the blast furnace in the 50th frame when it is not charged, Figure (c) is the thermal imaging image of the top of the blast furnace in the 60th frame when it is not charged, Figure (d) is the thermal imaging image of the top of the blast furnace in the 70th frame when it is not charged, Figure (e) is the thermal imaging image of the top of the blast furnace in the 80th frame when it is not charged, and Figure (f) is the thermal imaging image of the top of the blast furnace in the 100th frame when it is not charged.

[0051] Figure 3 These are thermal imaging images during the charging period in an embodiment of the present invention, wherein Figure (a) is a thermal imaging image of the top of the blast furnace in the first frame during the charging period, Figure (b) is a thermal imaging image of the top of the blast furnace in the 50th frame during the charging period, Figure (c) is a thermal imaging image of the top of the blast furnace in the 60th frame during the charging period, Figure (d) is a thermal imaging image of the top of the blast furnace in the 70th frame during the charging period, Figure (e) is a thermal imaging image of the top of the blast furnace in the 80th frame during the charging period, and Figure (f) is a thermal imaging image of the top of the blast furnace in the 100th frame during the charging period;

[0052] Figure 4 This is a texture feature map of the unloaded stage in an embodiment of the present invention;

[0053] Figure 5 This is a representative sample image randomly selected during the unloaded stage in an embodiment of the present invention;

[0054] Figure 6 This is a texture feature map of the loading stage in an embodiment of the present invention;

[0055] Figure 7 This is a representative sample image randomly selected during the loading stage in an embodiment of the present invention;

[0056] Figure 8 This is a schematic diagram of the final temperature field in an embodiment of the present invention;

[0057] Figure 9 This is a flowchart of the shared ICA-PCA algorithm in an embodiment of the present invention;

[0058] Figure 10 This is a diagram showing the proportion of carbon dioxide in the gaseous composition in an embodiment of the present invention;

[0059] Figure 11 This is a diagram showing the proportion of carbon monoxide in the gaseous composition in an embodiment of the present invention;

[0060] Figure 12 This is a distribution diagram of the carbon dioxide (CO2) prediction error in an embodiment of the present invention;

[0061] Figure 13 This is a distribution diagram of carbon dioxide (CO) prediction error in an embodiment of the present invention. Detailed Implementation

[0062] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0063] Blast furnace ironmaking is crucial in steel production. The blast furnace has a complex internal structure, operates under high temperature and pressure, and exhibits strong coupling and significant hysteresis. It involves multiple factors that directly or indirectly affect carbon emissions, such as temperature gradients, mass flow, and thermodynamic balance. These factors interact, directly influencing the formation and distribution of carbon emissions. Therefore, a thorough understanding of the blast furnace's internal structure and production process is essential before establishing a carbon emission prediction model to create an accurate prediction model and achieve precise control.

[0064] The main components of blast furnace emissions include top gas components such as hydrogen, nitrogen, carbon monoxide, and carbon dioxide. Hydrogen mainly acts as a reducing agent in the reaction, but its concentration is relatively low. Nitrogen mainly plays a dilution role and has the least direct impact on the blast furnace reaction process. Therefore, the focus is more on the concentrations of carbon monoxide and carbon dioxide and their impact on blast furnace operation.

[0065] Carbon monoxide (CO) is a key reducing agent in the iron ore reduction process in blast furnaces, increasing the reduction rate and iron production. However, excessively high CO concentrations may indicate incomplete combustion, leading to increased coke consumption and potential safety hazards. CO2 is the main product of coke combustion, and its concentration reflects combustion conditions. High CO2 concentrations indicate complete combustion and improved thermal efficiency, but excessively high CO2 levels can lead to uneven heat distribution, affecting thermodynamic equilibrium and the iron ore reduction process in the blast furnace.

[0066] The structure of the blast furnace, the generation and treatment of oxide monoxide and carbon monoxide are as follows:

[0067] 1. Feeding system: Raw materials such as ore and coke are fed into the blast furnace through the feeding system to ensure a continuous supply of materials.

[0068] 2. Top feeding: The material enters the top of the furnace and gradually descends due to gravity.

[0069] 3. Blast furnace body:

[0070] Throat and Shaft: In the throat and shaft, the raw materials begin to heat up. During this preheating stage, no significant chemical reactions occur.

[0071] Abdomen and waist: As the material descends further, the temperature rises, and the coke begins to burn, producing carbon dioxide and releasing a large amount of heat. The main chemical reactions here are:

[0072] C + O2 → CO2;

[0073] 4. Reduction reaction: As the temperature continues to decrease and increase, carbon dioxide reacts with coke to produce carbon monoxide:

[0074] CO2 + 2C → 2CO;

[0075] This endothermic reaction occurs in high-temperature areas, especially in the lower and middle sections of the furnace.

[0076] 5. Hot air system: Hot air is blown into the bottom of the blast furnace through the tuyeres, where it undergoes violent combustion with the descending coke, producing a large amount of carbon monoxide and carbon dioxide, thereby increasing the temperature inside the furnace.

[0077] 6. Reaction process: Carbon monoxide reacts with iron ore (iron oxide) in the furnace, reducing it to iron and generating CO.

[0078] 3CO + Fe₂O₃ → 2Fe + 3CO₂;

[0079] The generated carbon dioxide continues to react with coke to produce more carbon dioxide.

[0080] 7. Blast Furnace Gas Treatment System and Gas Recovery Device: The blast furnace gas treatment system treats and recovers gas from the furnace through gravity dust collectors, silencers, and water seal devices. The treated blast furnace gas can be used for power generation or as fuel.

[0081] Accurate prediction of CO and CO2 concentrations in blast furnace emissions is crucial for the safety, environmental protection, and economic efficiency of blast furnace operations. By monitoring and predicting these gas concentrations in real time, operators can adjust operating parameters promptly to achieve optimal control, reduce coke consumption, lower production costs, effectively reduce CO2 emissions, avoid safety risks, and ensure the safe operation of the blast furnace.

[0082] Based on this, the present invention provides a method for predicting blast furnace carbon emission concentration based on digital twins, combined with Figure 1 This may include the following steps:

[0083] Step 1: Real-time acquisition of production data during the blast furnace smelting process. This production data includes operating parameters, tuyere parameters, and multiple thermal images of the blast furnace top. The tuyere parameters include tuyere temperature and tuyere velocity. Figure 1 Actual production data of medium and high blast furnace smelting;

[0084] The operating parameters include 24 parameters such as oxygen concentration, central feed rate, injection intensity, and furnace static pressure.

[0085] The experimental data for this invention comes from actual production data of blast furnace No. 7 at a domestic steel plant. A flame video acquisition module was installed inside the blast furnace to obtain thermal images of the blast furnace top during actual production. The video frame rate was 12.5 frames per second, and each frame was a 384×288 RGB three-channel color image. Simultaneously with video acquisition, the computer recorded and saved the corresponding blast furnace production and operation parameters to a database.

[0086] Because the collected thermal images of the blast furnace top show instances where the flames are obscured by the blast furnace's charging chute, such as... Figure 2 Figure (c) in the middle, Figure 3 Figure (c) in the middle, where, Figure 2 These are thermal images of the blast furnace without charging. Among them, Figure (a) is the thermal image of the top of the blast furnace in frame 1 without charging, Figure (b) is the thermal image of the top of the blast furnace in frame 50 without charging, Figure (c) is the thermal image of the top of the blast furnace in frame 60 without charging, Figure (d) is the thermal image of the top of the blast furnace in frame 70 without charging, Figure (e) is the thermal image of the top of the blast furnace in frame 80 without charging, and Figure (f) is the thermal image of the top of the blast furnace in frame 100 without charging. Figure 3 These are thermal images during the charging period, where Figure (a) is the thermal image of the top of the blast furnace in frame 1 during the charging period, Figure (b) is the thermal image of the top of the blast furnace in frame 50 during the charging period, Figure (c) is the thermal image of the top of the blast furnace in frame 60 during the charging period, Figure (d) is the thermal image of the top of the blast furnace in frame 70 during the charging period, Figure (e) is the thermal image of the top of the blast furnace in frame 80 during the charging period, and Figure (f) is the thermal image of the top of the blast furnace in frame 100 during the charging period.

[0087] If the flames in the thermal imaging are obscured by the charging chute of the blast furnace, the accuracy of subsequent processing will be reduced. Therefore, it is necessary to filter the thermal imaging of the top of the blast furnace, which is achieved through step 2.

[0088] Step 2: Filter all thermal images of the top of the blast furnace to obtain thermal images that are not obscured by the charging chute;

[0089] By using the gray-level co-occurrence matrix, feature extraction is performed on each thermal image of the top of the blast furnace to obtain the texture features of each thermal image of the top of the blast furnace. The texture features include contrast, energy, uniformity, and pixel difference in the thermal image of the top of the blast furnace. Based on the thermal image of the top of the blast furnace and the corresponding texture features, all thermal images of the top of the blast furnace are filtered to obtain thermal images that are not obscured by the charging chute.

[0090] Specifically, based on the thermal imaging images of the blast furnace top and their corresponding texture features, all thermal imaging images of the blast furnace top were filtered to obtain those not obscured by the charging chute, including:

[0091] Based on the thermal imaging image of the blast furnace top, the production stage of the blast furnace is determined. When the blast furnace is in the uncharged stage, the range of texture feature values ​​that are not obscured by the charging chute is obtained and used as the target range. When the blast furnace is in the charging stage, the range of texture feature values ​​that are not obscured by the charging chute is obtained and used as the target range. A thermal imaging image of the blast furnace top with texture features within the target range is obtained to obtain a thermal imaging image that is not obscured by the charging chute.

[0092] In the specific implementation process, the texture features in the unloaded stage are as follows: Figure 4 Of these, 7 images met the requirements. A representative sample, frame 34, was randomly selected. Figure 5 As shown. The texture features during the loading stage are as follows. Figure 6 Of these, 5 images met the requirements. A representative sample, frame 83, was randomly selected, as shown below. Figure 7 As shown.

[0093] Step 3: Based on the thermal imaging image and vent parameters that are not obstructed by the loading chute, obtain the temperature field characteristics, which include the temperature gradient, maximum temperature and average temperature.

[0094] Step 3 specifically includes:

[0095] The mass conservation equation, momentum conservation equation, and energy conservation equation are used to process the vent parameters to obtain the initial temperature field. For the flame area in the thermal image not obscured by the loading chute, the grayscale value of the flame area is extracted, and the corresponding actual temperature value is obtained. The actual temperature value is then combined with the initial temperature field to obtain the final temperature field, such as... Figure 8 The final temperature field is then subjected to feature extraction to obtain temperature field features.

[0096] In practical implementation, the energy conservation equation needs to be simplified to obtain the simplified energy conservation equation, which can be expressed by the following formula:

[0097]

[0098] The mass conservation equation is expressed by the following formula:

[0099]

[0100] The momentum conservation equation is expressed by the following formula:

[0101]

[0102] Step 4: Combining Figure 9For each thermal image not obscured by the loading chute, features are extracted using ResNet50 and an autoencoder to obtain image features for all unobstructed thermal images. These features are then compressed into an image feature matrix X. Independent component analysis (ICA) is used to extract non-Gaussian independent source features S from the image feature matrix X. Inverse ICA transform is then applied to process the non-Gaussian independent source features S to obtain the key image features X. e X, the key features of the image e Combined with the physical variable Z, the shared matrix X is obtained. c =[Z;X e ], where the physical variable Z includes the preprocessed operating parameters and temperature field characteristics, and the principal component analysis algorithm is used to extract Gaussian features from the shared matrix; where the operating parameters are Figure 1 The operating parameters in the data, after data preprocessing, yield parameters such as oxygen enrichment rate, central ore addition amount, and blower kinetic energy, which are the preprocessed operating parameters.

[0103] The independent component analysis algorithm is used to extract the non-Gaussian independent source features S from the image feature matrix X, which is achieved through the following formula:

[0104] S = WX;

[0105] Where W is the unmixing matrix.

[0106] Specifically, the inverse ICA transform is used to process the non-Gaussian independent source features S to obtain the key image features X. e Specifically, this is achieved through the following formula:

[0107] X e＝ X-AS = (I-AW)X;

[0108] Where A is the mixing coefficient matrix and I is the identity matrix.

[0109] In the specific implementation process, the cumulative variance contribution rate of the principal components is shown in Table 1:

[0110] Table 1. Cumulative Variance Contribution of Principal Components

[0111] Main ingredients Variance contribution rate (%) PCA 1 33.38 PCA 2 29.41 PCA 3 10.90 PCA4 6.16 PCA 5 6.02 PCA6 4.66 PCA 7 3.49 PCA 8 1.81 Cumulative variance contribution (%) 95.82

[0112] We use Gaussian features obtained from multi-source data through dimensionality reduction via Principal Component Analysis (PCA) and non-Gaussian features extracted from image data through Independent Component Analysis (ICA) as inputs to the prediction model. To improve prediction accuracy, we employ an optimization algorithm to determine the weights of each model, thereby achieving accurate prediction of the CO2 and CO ratios in blast furnace carbon emissions.

[0113] Step 5: Obtain the final carbon monoxide and carbon dioxide concentrations based on the non-Gaussian independent source characteristic S and the Gaussian characteristic.

[0114] Step 5 specifically includes:

[0115] Step 5.1: Using the first prediction model based on the random forest regression algorithm, process the non-Gaussian independent source feature S in Step 4 to obtain the first concentration prediction values ​​of carbon monoxide and carbon dioxide.

[0116] The first prediction model is obtained by training multiple first training samples. The first training samples include first input samples and first output samples. The first input samples are historical non-Gaussian independent source features. The first output samples include the true values ​​of carbon monoxide concentration and carbon dioxide concentration corresponding to the historical non-Gaussian independent source features.

[0117] Random Forest (RF) is an ensemble learning algorithm (see "Random Forests") that uses multiple decision trees to make predictions and aggregates their results to obtain a final prediction. This method addresses the overfitting problem associated with individual decision trees, thereby improving the model's accuracy and generalization ability. Random Forest performs exceptionally well in handling high-dimensional, complex data and is well-suited for predicting blast furnace carbon emission concentrations.

[0118] Traditional neural network models perform well in handling complex and highly uncertain objects, but training is often time-consuming and prone to problems such as local optima, overfitting, and poor robustness. However, carbon emission prediction models require real-time prediction results. To address these issues, Bao et al. proposed the Random Weighted Neural Network (RVFLN) algorithm. RVFLNs are feedforward neural networks with only one hidden layer. The weights and biases from the input layer to the hidden layer are randomly generated within a certain range, while the weights from the hidden layer to the output layer are solved using the least squares method, avoiding the iterative process. Therefore, RVFLNs have the advantages of short training time and strong generalization ability, as specifically achieved in step 5.2.

[0119] Step 5.2: The Gaussian features from Step 4 are processed using multiple different second prediction models based on the random weighted neural network algorithm to obtain multiple second prediction values. These second prediction values ​​include the second concentration prediction values ​​for carbon monoxide and carbon dioxide. Based on preset weights, a weighted average of all the second concentration prediction values ​​for carbon monoxide is calculated to obtain the third concentration prediction value for carbon monoxide. Similarly, a weighted average of all the second concentration prediction values ​​for carbon dioxide is calculated to obtain the third concentration prediction value for carbon dioxide. For details on the random weighted neural network algorithm, please refer to the reference "Functional-LinkNet Computing: Theory, System Architecture, and Functionalities".

[0120] Step 5.3: Using the Particle Swarm Optimization (PSO) algorithm, determine the first and second weights. Based on these weights, perform a weighted sum of the first and third predicted carbon monoxide concentrations to obtain the final carbon monoxide concentration. Similarly, perform a weighted sum of the first and third predicted carbon dioxide concentrations to obtain the final carbon dioxide concentration. The final carbon monoxide concentration is... Figure 1 The CO content in the exhaust gas from the blast furnace, and the final carbon dioxide concentration. Figure 1 CO2 content in the exhaust gas from medium and high blast furnaces.

[0121] In step 5.3, multiple different second prediction models are obtained through the following steps:

[0122] Step A1: Obtain historical Gaussian features and the corresponding true concentration values, including the true concentration values ​​of carbon monoxide and carbon dioxide. Use the historical Gaussian features as the second input sample and the true concentration values ​​as the second output sample. The second input sample and the second output sample form the second training sample, thereby obtaining multiple second training samples.

[0123] Step A2: Using a random sampling strategy with replacement, Bagging sampling is performed on multiple second training samples to obtain multiple second training samples, forming a subset of training sets, and thus obtaining multiple subsets of training sets.

[0124] Step A3: For each sub-training set, construct an initial model of a random weighted neural network algorithm, wherein the weights and biases of the initial model are random;

[0125] Step A4: Train the initial model using a sub-training set. Specifically, input samples from the sub-training set are input into the initial model to obtain historical predicted values ​​corresponding to historical Gaussian features. These historical predicted values ​​include predicted values ​​for carbon monoxide concentration and carbon dioxide concentration. Based on the historical predicted values ​​and output samples, a regularized least squares algorithm is used to update the parameters in the initial model until the model converges, resulting in a second prediction model. This process leads to multiple different second prediction models corresponding to multiple sub-training sets.

[0126] Step 5.3 uses the Particle Swarm Optimization (PSO) algorithm to determine the first and second weights, including:

[0127] Step B1: Obtain the predicted value and the actual value of carbon monoxide concentration, as well as the predicted value and the actual value of carbon dioxide concentration. The predicted value of carbon monoxide concentration includes a first concentration value and a second concentration value, and the predicted value of carbon dioxide concentration includes a third concentration value and a fourth concentration value.

[0128] Step B2: Initialize the velocity and position of all particles in the particle swarm. The position of each particle includes two randomly initialized weights. Initialize the individual optimal position and the global optimal position, and set the initial number of iterations.

[0129] Step B3: Take the initial velocity of the particle as the current velocity, the initial position of the particle as the current position, the initial individual optimal position of each particle as the current individual optimal position of that particle, the initial global optimal position as the current global optimal position, and the initial number of iterations as the current number of iterations;

[0130] Step B4: For each particle, according to the weight in the particle's current position, the first concentration value and the second concentration value are weighted and summed to obtain the weighted predicted value of carbon monoxide concentration; the third concentration value and the fourth concentration value are weighted and summed to obtain the weighted predicted value of carbon dioxide concentration.

[0131] Step B5: For each particle, calculate the mean square error between the weighted predicted value of carbon monoxide concentration and the actual value of carbon monoxide concentration to obtain the first error, and calculate the mean square error between the weighted predicted value of carbon dioxide concentration and the actual value of carbon dioxide concentration to obtain the second error. The first error and the second error constitute the current fitness of the particle. Among the current fitness of the particle and the fitness corresponding to the current individual optimal position of the particle, select the position with the smaller fitness as the new current individual optimal position of the particle.

[0132] Step B6: Among the fitness of the current individual best position after all particles are updated, and the fitness of the current global best position, select the position with the smaller fitness as the new current global best position;

[0133] Step B7: Update the current position and current velocity of each particle, specifically using the following formula:

[0134] v i,t+1 =ωv i,t +c1r1(p best,i -x i,t )+c2r2(g best -x i,t );

[0135] x i,t+1 =x i,t +v i,t+1 ;

[0136] Where ω is the inertia weight, c1 and c2 are the acceleration constants, r1 and r2 are random numbers, and v i,t+1 Let v be the current velocity of the i-th particle at iteration number t+1. i,t Let p be the current velocity of the i-th particle at the current iteration number t. best,i Let g be the optimal position of the i-th particle in the current individual at the current iteration number t. best Let x be the current globally optimal position of the i-th particle at the current iteration number t. i,t Let x be the current position of the i-th particle at the current iteration number t. i,t+1 Let be the current position of the i-th particle at iteration number t+1;

[0137] Step B8: Increment the current iteration count by one. If the current iteration count is less than the preset count, or the fitness of the new current global optimal position is less than the preset threshold, use the two weights in the new current global optimal position as the first weight and the second weight. If the current iteration count is not less than the preset count, and the fitness of the new current global optimal position is not less than the preset threshold, return to step B4 until the current iteration count is less than the preset count, or the fitness of the new current global optimal position is less than the preset threshold.

[0138] Based on the above technical solution, this invention underwent experimental verification. The invention uses blast furnace No. 7 of a domestic steel plant as the research object and utilizes actual production record data. The proposed RF-RVFLNs combined model was verified through on-site video images, actual blast furnace operating parameters, and the ratio of CO2 to CO in the blast furnace exhaust gas. The preprocessed dataset of 1608 samples was divided into training and testing sets, with 1408 samples used for training and 200 for testing. The ratios of CO2 and CO in the blast furnace exhaust gas used for testing the model are shown below. Figure 10 and Figure 11As shown. This invention also explored other methods, such as deep learning-based mechanism fusion methods, to predict blast furnace carbon emission concentrations, but these methods were not as accurate as the method proposed in this invention.

[0139] To more intuitively demonstrate the higher accuracy and effectiveness of the proposed combined model, this study uses test set data to compare the prediction results of individual models and the combined model. The selected comparison algorithms include Multi-Kernel Support Vector Regression (M-SVR), Sparse Autoencoder (S-DAE), Random Forest (RF), and Random Vector Function Linked Networks (RVFLNs). These algorithms were used to predict data without furnace top images. The CO2 prediction error is shown below. Figure 12 As shown, the CO prediction error is as follows: Figure 13 As shown.

[0140] Root mean square error (RMSE) and mean absolute percentage error (MAPE) have been used as evaluation metrics, specifically calculated using the following formula:

[0141]

[0142] Table 2 lists the comparison results of RMSE and MAPE of different algorithms in CO2 emission concentration prediction, and Table 3 lists the comparison results of RMSE and MAPE in CO emission concentration prediction.

[0143] Table 2 Comparison of RMSE and MAPE results of different algorithms in CO2 emission concentration prediction.

[0144] algorithm RMSE (%) MAPE (%) M-SVR 0.6839 2.7921 S-DAE 0.8002 3.1857 RF 0.2956 1.0332 RVFLNS 0.3374 1.7590 RF-RVFLNS 0.2174 0.8742

[0145] Table 3 Comparison of RMSE and MAPE in CO emission concentration prediction.

[0146] algorithm RMSE (%) MAPE (%) M-SVR 0.7626 2.1964 S-DAE 0.6452 1.7748 RF 0.3383 0.9311 RVFLNS 0.3716 1.4461 RF-RVFLNS 0.2844 0.8396

[0147] In summary, this invention utilizes the Gray-Level Co-occurrence Matrix (GLCM) to select unobstructed flame portions from top thermal images and extracts features from these images using ResNet50 and an autoencoder. Then, Independent Component Analysis (ICA) is applied to extract non-Gaussian independent source features from the image data, which serve as input to a Random Forest (RF) model. Next, a mechanistic model is used to construct the temperature field within the blast furnace. Key features of the images are reconstructed using inverse ICA transform and combined with operating parameters and temperature field features to establish a feature-sharing matrix. Finally, Principal Component Analysis (PCA) is used to extract principal components, which serve as input to a Random Vector Functional Linked Network (RVFLN) ensemble. Finally, Particle Swarm Optimization (PSO) is used to fuse the output results. The results show that the proposed model has a mean absolute percentage error (MAPE) of 0.8742% for CO2 concentration and 0.8396% for CO2 concentration, demonstrating its high accuracy and practical application value.

[0148] This invention proposes a carbon emission concentration prediction model based on digital twin technology, which combines video images, operating parameters, and temperature field features. By employing Independent Component Analysis (ICA) and Principal Component Analysis (PCA) algorithms for feature fusion and using a Random Forest (RF)-Extreme Learning Machine (RVFLNs) ensemble model for prediction, experimental results show that this method significantly improves the accuracy and stability of carbon emission prediction in real-world production environments. It effectively utilizes the strengths of each model, overcoming the limitations of a single model. However, this research still has some limitations. During image feature extraction, some local features may be lost. Future research could explore how to directly use the image matrix for regression learning while maintaining real-time performance.

[0149] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that this disclosure should also cover other technical solutions formed by any combination of the above-described technical features or their equivalents without departing from the inventive concept described above. For example, the scope of the invention involved in the disclosed embodiments is not limited to the technical solutions formed by specific combinations of the above-described technical features, but rather by substituting the above-described features with technical features that have similar functions to those disclosed in the embodiments of this disclosure.

Claims

1. A method for predicting carbon emission concentration from blast furnaces based on digital twins, characterized in that, include: Step 1: Real-time acquisition of production data during the blast furnace smelting process. The production data includes operating parameters, tuyere parameters, and multiple thermal imaging images of the top of the blast furnace. The tuyere parameters include tuyere temperature and tuyere velocity. Step 2: Filter all thermal images of the top of the blast furnace to obtain thermal images that are not obscured by the charging chute; Step 3: Based on the thermal imaging image and vent parameters that are not obstructed by the loading chute, obtain the temperature field characteristics, which include the temperature gradient, maximum temperature and average temperature. Step 4: Extract features from each thermal image that is not obstructed by the loading chute using ResNet50 and an autoencoder to obtain the image features of all thermal images that are not obstructed by the loading chute, and compress them into an image feature matrix. X Independent component analysis algorithm is used to analyze the image feature matrix. X Extraction is performed to obtain non-Gaussian independent source features. S The inverse ICA transform is used to analyze the features of non-Gaussian independent sources. S The image is processed to obtain key features. X e Image key features X e Combined with the physical variable Z, a shared matrix is ​​obtained. X c =[ Z ; X e ], where the physical variable Z includes the preprocessed operating parameters and temperature field characteristics, and the principal component analysis algorithm is used to extract the Gaussian features in the shared matrix; Step 5: Based on the characteristics of non-Gaussian independent sources S Using Gaussian characteristics, the final carbon monoxide and carbon dioxide concentrations are obtained; Step 5.1: Using the first prediction model based on the random forest regression algorithm, analyze the non-Gaussian independent source features from Step 4. S The process yields the first predicted concentration values ​​for carbon monoxide and carbon dioxide. The first prediction model is obtained by training multiple first training samples. The first training samples include first input samples and first output samples. The first input samples are historical non-Gaussian independent source features. The first output samples include the true values ​​of carbon monoxide concentration and carbon dioxide concentration corresponding to the historical non-Gaussian independent source features. Step 5.2: The Gaussian features in Step 4 are processed by multiple different second prediction models based on the random weighted neural network algorithm to obtain multiple second prediction values. The second prediction values ​​include the second concentration prediction values ​​of carbon monoxide and carbon dioxide. According to the preset weights, the second concentration prediction values ​​of all carbon monoxide are weighted and averaged to obtain the third concentration prediction value of carbon monoxide. The second concentration prediction values ​​of all carbon dioxide are weighted and averaged to obtain the third concentration prediction value of carbon dioxide. Step 5.3: Using the Particle Swarm Optimization (PSO) algorithm, determine the first and second weights. Based on the first and second weights, perform a weighted sum of the first and third predicted carbon monoxide concentrations to obtain the final carbon monoxide concentration. Similarly, based on the first and second weights, perform a weighted sum of the first and third predicted carbon dioxide concentrations to obtain the final carbon dioxide concentration.

2. The method for predicting blast furnace carbon emission concentration based on digital twins according to claim 1, characterized in that, Step 2 specifically includes: By using the gray-level co-occurrence matrix, feature extraction is performed on each thermal image of the top of the blast furnace to obtain the texture features of each thermal image of the top of the blast furnace. The texture features include contrast, energy, uniformity, and pixel difference in the thermal image of the top of the blast furnace. Based on the thermal image of the top of the blast furnace and the corresponding texture features, all thermal images of the top of the blast furnace are filtered to obtain thermal images that are not obscured by the charging chute.

3. The method for predicting blast furnace carbon emission concentration based on digital twins according to claim 2, characterized in that, Based on the thermal imaging images of the blast furnace top and their corresponding texture features, all thermal imaging images of the blast furnace top were filtered to obtain those not obscured by the charging chute, including: Based on the thermal imaging image of the blast furnace top, the production stage of the blast furnace is determined. When the blast furnace is in the uncharged stage, the range of texture feature values ​​that are not obscured by the charging chute is obtained and used as the target range. When the blast furnace is in the charging stage, the range of texture feature values ​​that are not obscured by the charging chute is obtained and used as the target range. A thermal imaging image of the blast furnace top with texture features within the target range is obtained to obtain a thermal imaging image that is not obscured by the charging chute.

4. The method for predicting blast furnace carbon emission concentration based on digital twins according to claim 1, characterized in that, Step 3 specifically includes: The mass conservation equation, momentum conservation equation, and energy conservation equation are used to process the vent parameters to obtain the initial temperature field. For the flame area in the thermal imaging image that is not blocked by the loading chute, the gray value of the flame area is extracted, and the actual temperature value corresponding to the gray value is obtained. The actual temperature value is combined with the initial temperature field to obtain the final temperature field. The final temperature field is then subjected to feature extraction to obtain the temperature field features.

5. The method for predicting blast furnace carbon emission concentration based on digital twins according to claim 1, characterized in that, In step 5.3, multiple different second prediction models are obtained through the following steps: Step A1: Obtain historical Gaussian features and the corresponding true concentration values, including the true concentration values ​​of carbon monoxide and carbon dioxide. Use the historical Gaussian features as the second input sample and the true concentration values ​​as the second output sample. The second input sample and the second output sample form the second training sample, thereby obtaining multiple second training samples. Step A2: Using a random sampling strategy with replacement, Bagging sampling is performed on multiple second training samples to obtain multiple second training samples, forming a subset of training sets, and thus obtaining multiple subsets of training sets. Step A3: For each sub-training set, construct an initial model of a random weighted neural network algorithm, wherein the weights and biases of the initial model are random; Step A4: Train the initial model using a sub-training set. Specifically, input samples from the sub-training set are input into the initial model to obtain historical predicted values ​​corresponding to historical Gaussian features. These historical predicted values ​​include predicted values ​​for carbon monoxide concentration and carbon dioxide concentration. Based on the historical predicted values ​​and output samples, a regularized least squares algorithm is used to update the parameters in the initial model until the model converges, resulting in a second prediction model. This process leads to multiple different second prediction models corresponding to multiple sub-training sets.

6. The method for predicting blast furnace carbon emission concentration based on digital twins according to claim 1, characterized in that, Step 5.3 uses the Particle Swarm Optimization (PSO) algorithm to determine the first and second weights, including: Step B1: Obtain the predicted value and the actual value of carbon monoxide concentration, as well as the predicted value and the actual value of carbon dioxide concentration. The predicted value of carbon monoxide concentration includes a first concentration value and a second concentration value, and the predicted value of carbon dioxide concentration includes a third concentration value and a fourth concentration value. Step B2: Initialize the velocity and position of all particles in the particle swarm. The position of each particle includes two randomly initialized weights. Initialize the individual optimal position and the global optimal position, and set the initial number of iterations. Step B3: Take the initial velocity of the particle as the current velocity, the initial position of the particle as the current position, the initial individual optimal position of each particle as the current individual optimal position of that particle, the initial global optimal position as the current global optimal position, and the initial number of iterations as the current number of iterations; Step B4: For each particle, according to the weight in the particle's current position, the first concentration value and the second concentration value are weighted and summed to obtain the weighted predicted value of carbon monoxide concentration; the third concentration value and the fourth concentration value are weighted and summed to obtain the weighted predicted value of carbon dioxide concentration. Step B5: For each particle, calculate the mean square error between the weighted predicted value of carbon monoxide concentration and the actual value of carbon monoxide concentration to obtain the first error, and calculate the mean square error between the weighted predicted value of carbon dioxide concentration and the actual value of carbon dioxide concentration to obtain the second error. The first error and the second error constitute the current fitness of the particle. Among the current fitness of the particle and the fitness corresponding to the current individual optimal position of the particle, select the position with the smaller fitness as the new current individual optimal position of the particle. Step B6: Among the fitness of the current individual best position after all particles are updated, and the fitness of the current global best position, select the position with the smaller fitness as the new current global best position; Step B7: Update the current position and current velocity of each particle, specifically using the following formula: ； in, For inertial weights, , The acceleration constant, , It is a random number. For the i-th particle at iteration number t +1 to the current speed, For the i-th particle in the current iteration number t The current speed, For the i-th particle in the current iteration number t The current optimal position of the individual. For the i-th particle in the current iteration number t The current global optimal position, For the first i The particles at the current iteration number t Current location For the i-th particle at iteration number t +1 Current position; Step B8: Increment the current iteration count by one. If the current iteration count is less than the preset count, or the fitness of the new current global optimal position is less than the preset threshold, use the two weights in the new current global optimal position as the first weight and the second weight. If the current iteration count is not less than the preset count, and the fitness of the new current global optimal position is not less than the preset threshold, return to step B4 until the current iteration count is less than the preset count, or the fitness of the new current global optimal position is less than the preset threshold.

7. The method for predicting blast furnace carbon emission concentration based on digital twins according to claim 1, characterized in that, Independent component analysis algorithm is used to analyze the image feature matrix. X Extraction is performed to obtain non-Gaussian independent source features. S Specifically, this is achieved through the following formula: S=WX ； in, W This is the unmixing matrix.

8. The method for predicting blast furnace carbon emission concentration based on digital twins according to claim 1, characterized in that, The inverse ICA transform is used to analyze the features of non-Gaussian independent sources. S The image is processed to obtain key features. X e Specifically, this is achieved through the following formula: X e =X-AS= ( I-AW ) X ； in, A The mixing coefficient matrix, I It is an identity matrix.

Images