Hydrogen-based shaft furnace one-dimensional multi-phase field modeling method based on fuzzy neural network optimization

Abstract

The invention discloses a hydrogen-based shaft furnace one-dimensional multi-phase field modeling method based on fuzzy neural network optimization. The method comprises the following steps: firstly, constructing a multi-phase field one-dimensional preliminary model of internal temperature, pressure and concentration of the hydrogen-based shaft furnace by using a fluid mechanics equation; secondly, constructing a reaction model based on an additive reaction time law, endowing each characteristic time item with a coefficient multiplier, and fully exciting the model through random values of the coefficient multipliers to obtain multi-phase field model simulation data; then constructing an agent model and optimizing a coefficient multiplier by utilizing a fuzzy neural network and combining a non-dominated sorting genetic algorithm II (NSGA-II); and finally, inputting the optimized coefficient multiplier and newly collected shaft furnace feeding parameters into the multi-phase field model to obtain an optimized hydrogen-based shaft furnace multi-phase field simulation result. Through cooperative driving of a mechanism and data, precise modeling of a multi-phase field in the hydrogen-based shaft furnace is achieved, and high simulation precision is achieved for different smelting conditions.

Description

Technical Field

[0001] This invention belongs to the field of multiphase field modeling inside hydrogen-based vertical shaft furnaces, and relates to a one-dimensional multiphase field modeling method for hydrogen-based vertical shaft furnaces based on fuzzy neural network optimization. Background Technology

[0002] Hydrogen-based shaft furnaces, as a key innovative device in the ironmaking industry, operate using hydrogen as the primary reducing agent, aiming to significantly reduce carbon dioxide emissions. The traditional steel industry, based on blast furnace ironmaking, accounts for approximately 6% of global carbon emissions, with the blast furnace ironmaking process itself accounting for over 70% of that. Therefore, in-depth research on hydrogen-based shaft furnaces to replace traditional blast furnace ironmaking processes is of great significance. Constructing a multiphase field model of the internal structure of a hydrogen-based shaft furnace allows for a deeper understanding of its complex physicochemical processes, including gas diffusion, convection transport, reduction reactions, and particle flow, thereby optimizing furnace operation and improving system efficiency and environmental benefits.

[0003] Existing modeling methods can be broadly categorized into two types: multiphysics coupling mechanism modeling and data-driven modeling. Multiphysics coupling mechanism modeling can be further divided into lumped parameter models and distributed parameter models. Lumped parameter models divide the shaft furnace into several reaction blocks and use a limited number of state variables (such as temperature, pressure, and chemical composition) to describe the average behavior of the entire system, offering the advantage of high computational efficiency. However, neglecting the spatial distribution characteristics of physical quantities within the shaft furnace makes it difficult for lumped parameter models to reflect the detailed characteristics of the multiphase fields inside the furnace, such as the gradient distribution of chemical reaction rates and the non-uniformity of the gas flow field.

[0004] Distributed parameter models provide a more detailed spatial description of the system, using transfer equations (such as mass conservation equations, energy conservation equations, and momentum conservation equations) to describe the spatial distribution characteristics within the shaft furnace. While suitable for describing multiphase flow and heat transfer processes within shaft furnaces, the distributed parameter method suffers from high modeling complexity. Furthermore, the excessive assumptions made during modeling make it prone to deviations from reality, making timely correction through real-time acquisition of industrial process data difficult. Over long periods of operation, distributed parameter models struggle to maintain optimal performance.

[0005] With the rapid development of technologies for industrial field data acquisition, transmission, and storage, data is playing an increasingly important role in the modeling process, thus data-driven vertical shaft furnace modeling has gradually attracted attention. This method does not delve into the dynamic evolution mechanism inside the vertical shaft furnace, but relies solely on large-scale real-time operational data as input and output, combined with machine learning and artificial intelligence technologies to achieve dynamic modeling and prediction of the system. However, modeling techniques relying solely on data have very high requirements for data quality, are highly susceptible to noise and outliers, and are prone to overfitting, resulting in poor model generalization and making it difficult to use for continuous modeling of real processes. Summary of the Invention

[0006] To overcome the shortcomings of existing modeling techniques, this invention provides a one-dimensional multiphase field modeling method for hydrogen-based vertical shaft furnaces based on fuzzy neural network optimization. This method can achieve accurate modeling of the multiphase field of hydrogen-based vertical shaft furnaces and has high simulation accuracy for different smelting conditions.

[0007] To achieve the above technical objectives, the one-dimensional multiphase field modeling method for hydrogen-based vertical shaft furnaces based on fuzzy neural network optimization includes the following steps:

[0008] First, a multiphase field model describing the distribution of gas and solid phase pressure, temperature, and component concentration inside a hydrogen-based vertical shaft furnace is constructed based on fluid dynamics equations. Second, a reaction model is constructed based on the additive reaction time law, and a coefficient multiplier is assigned to each characteristic time term in the model. The improved reaction model is used as the source term in the multiphase field model. Subsequently, multiple sets of feed and discharge parameters and key process variables of the vertical shaft furnace under real industrial conditions are extracted. The feed parameters of the vertical shaft furnace are combined with the randomly sampled coefficient multipliers and input into the constructed multiphase field model to obtain the corresponding multiphase field model solution. Then, the fuzzy neural network is trained using the simulation data of the multiphase field model, and the trained fuzzy neural network is optimized using the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) to obtain the optimal solution of the coefficient multipliers. Finally, the optimal solution of the coefficient multipliers and the key feed parameters of the vertical shaft furnace are re-input into the multiphase field model to generate the optimized multiphase field simulation results of the hydrogen-based vertical shaft furnace.

[0009] I. The construction of the multiphase field model based on the fluid dynamics equations includes the following steps:

[0010] (1) The momentum conservation process is described using the Ergun equation, which is expressed in the following form:

[0011]

[0012] Where the subscript g represents the gas phase; u g d represents the apparent gas flow rate, p represents the gas pressure, ε represents the porosity of the particle bed, and d represents the apparent gas flow rate. p Represents the diameter of ore particles, μ g and ρ g denoted by and , respectively, the viscosity and density of the gas phase, and z represents the axial direction of the vertical furnace.

[0013] Since the solid bed inside an actual hydrogen-based shaft furnace moves downwards slowly and almost no movement occurs in the cross-section, the solid velocity can be considered a fixed value, unaffected by factors such as pressure, and is expressed as:

[0014]

[0015] Where the subscript 's' represents the solid phase; W represents the mass flow rate of the solid; and ρ b S represents the apparent mass density of the bed, and S represents the cross-sectional area of ​​the vertical furnace.

[0016] (2) The concentration distribution of gas and solid components inside the hydrogen-based vertical furnace is described by a set of mass conservation equations, including the gas mass conservation equation, the reaction reduction degree distribution equation, and the relationship between the solid mass fraction and the reaction reduction degree. The gas mass conservation equation is expressed in the following form:

[0017]

[0018] The equation for the distribution of the degree of reduction of the reaction is expressed in the following form:

[0019]

[0020] In this context, the subscripts i and n represent the gas phase component and reaction number, respectively. i = H2 or H2O, and n = 1 to 3, representing the reactions 3Fe2O3 + H2 → 2Fe3O4 + H2O, Fe3O4 + H2 → 3FeO + H2O, and FeO + H2 → Fe + H2O, respectively; x i and X n R represents the mole fraction of gaseous component i and the degree of reaction of the nth reaction, respectively; n Represents the reaction rate of the nth reaction; n g Represents the gas phase molar flow rate; κ is the initial reducible oxygen content in a single particle; β is the initial oxygen molar fraction for each reduction step. n The values ​​are 0.1111, 0.1889, and 0.70, respectively.

[0021] Knowing the degree of reduction, the mass fraction of each solid component in the ore can be determined using the relationship between the solid mass fraction and the degree of reduction. This relationship is expressed as follows:

[0022]

[0023] in, w FeO w Fe The values ​​represent the mass fractions of the solid phase components Fe2O3, Fe3O4, FeO, and Fe, respectively.

[0024] (3) The temperature distribution of the gas and solid phases inside the hydrogen-based vertical furnace is described by a set of energy conservation equations, including the gas energy conservation equation and the solid energy conservation equation. The gas energy conservation equation is expressed in the following form:

[0025]

[0026] The solid energy conservation equation is expressed in the following form:

[0027]

[0028] Among them, T g and T s These represent the gas phase and solid phase temperatures, respectively; G represents the gas volumetric flow rate; C g and C s These represent the isobaric specific heat capacities of the gas phase and solid phase, respectively; h p and h w These represent the gas-solid heat transfer coefficient and the overall heat transfer coefficient between the gas and the outside environment through the water-cooled furnace wall, respectively; D represents the internal diameter of the vertical furnace; T a Represents ambient temperature; ΔH n This represents the enthalpy change of the nth reaction.

[0029] II. The construction of the reaction model based on the additive reaction time law includes the following steps:

[0030] (1) The reaction model includes three reactions: 3Fe₂O₃ + H₂ → 2Fe₃O₄ + H₂O, Fe₃O₄ + H₂ → 3FeO + H₂O, and FeO + H₂ → Fe + H₂O. Each reaction is divided into mass transfer stages and an analytical model is constructed using the additive reaction time law. The additive reaction time law states that when different transfer resistances (e.g., chemical reactions, diffusion, external mass transfer, etc.) occur in series, the total reaction time required to complete the complete conversion of the solid reactants can be approximated as the sum of the characteristic times of each individual step. The characteristic time of each individual step is the time required to achieve complete conversion in a system controlled only by that basic step. In the hydrogen-based vertical shaft furnace reaction model, these steps are: external mass transfer of gas through the boundary layer surrounding the particles, diffusion of gas through the pores of the particles, diffusion of gas within the grains of the particles, diffusion of gas within the crystals of the grains, and gas-solidification chemical reaction. The characteristic time of external mass transfer of gas through the boundary layer surrounding the particles is expressed as:

[0031]

[0032] The characteristic time of gas diffusion through particle pores is described as follows:

[0033]

[0034] The characteristic time of gas diffusion within the grains of a particle is described as follows:

[0035]

[0036] The characteristic time of gas diffusion within the crystal grain is described as follows:

[0037]

[0038] The characteristic time of gas-solidification chemical reactions is described as follows:

[0039]

[0040] Based on the characteristic time of each transmission resistance step, the formula for calculating the reaction rate is expressed as follows:

[0041]

[0042] Among them, c n k represents the concentration of the solid reactant in the nth reaction. eat Indicates the external mass transfer coefficient of the gas; These represent the effective diffusion coefficients of H2 within the particle, grain, and crystal, respectively; k ch,x and K n Let d represent the reaction rate constant and the reaction equilibrium constant of the nth reaction, respectively; gr and d cr These represent the diameters of the FeO grains and the crystals, respectively. and These represent the mole fractions of H2 and H2O outside the ore particles, respectively. This indicates the mole fraction of H2 at reaction equilibrium.

[0043] (2) Assign coefficient multipliers to each characteristic time term in the reaction model, as shown in the following formula:

[0044]

[0045] Where a, b, c, d, and e represent external mass transfer of gas through the boundary layer surrounding the particle, diffusion of gas through the pores of the particle, diffusion of gas within the grains of the particle, diffusion of gas within the crystals of the grains, and gas-solidification chemical reaction, respectively. The coefficient multiplier corresponding to the characteristic time term of external mass transfer of gas through the boundary layer surrounding the particle is used to adjust the weights of each term and eliminate the influence caused by deviations between the model parameter settings and the actual situation.

[0046] III. The extraction of key variables of the vertical shaft furnace and their combination with randomly sampled coefficient multipliers for input into the multiphase field model for simulation includes the following steps:

[0047] (1) Extract the feed and discharge parameters and key process variables of the vertical shaft furnace. The feed parameters of the vertical shaft furnace include the inlet gas velocity u of the H2 inlet pipe on the side of the vertical shaft furnace. g,in Intake pressure p in Intake air temperature T g,in The ore feed rate W at the top of the vertical shaft furnace and the feed temperature T s,in , Mass fraction of feed components w j,inBed porosity ε and apparent mass density ρ b Particle diameter d p The discharge parameters of the vertical shaft furnace include the gas temperature T at the top of the furnace. g,out Pressure p out Component x i,out Iron content w at the bottom of the furnace Fe,out Discharge temperature T s,out The key process variables include the gas temperature T at the furnace wall temperature sampling point. g,wall The subscripts in, out, and wall represent the values ​​of the variable at the corresponding feed, discharge, and furnace wall sampling points, respectively.

[0048] (2) The feed parameters of the vertical shaft furnace are combined with the randomly sampled coefficient multipliers and input into the constructed multiphase field model. Specifically, the feed parameter vector X = [u g,in ,p in ,T g,in ,W,T s,in ,w j,in ,ε,ρ b ,d p ] and the coefficient multipliers {θ1,θ2,…,θ} of k randomly sampled groups m ,…,θ k}(where θ) m =[a m ,b m ,c m ,d m ,e m The parameters m∈[1,k], k≥5) are combined to obtain the model input vector group {[X,θ1],[X,θ2],…,[X,θ2],[X,θ3],[X,θ4],[X,θ5],[X,θ6],[X,θ7],[X,θ8],[X,θ9],[X,θ1 ...2],[X,θ1],[X, k Then, input vectors [X, θ] to each model are used. m Input the multiphase field models constructed in step one and solve them to obtain the corresponding multiphase field model solution results.

[0049] The randomly sampled coefficient multipliers are specifically coefficient multiplier values ​​randomly sampled within a predetermined range [0.1, 10]. They are used to artificially correct for ore particles, grains, crystal diameters, and mass transfer and diffusion coefficients that are difficult to determine precisely in the reaction model. For different feed parameter vectors, the coefficient multiplier group needs to be re-randomized to ensure that all values ​​of the coefficient multipliers cover the range as much as possible, achieving a sufficient excitation effect.

[0050] IV. The optimization of coefficient multipliers based on multiphase field model simulation data using fuzzy neural networks and the NSGA-II algorithm includes the following steps:

[0051] (1) The output parameter vector Y = [T g,out ,p out ,x i,out ,w Fe,out ,T s,out ,T g,wall ] and the corresponding model input vector Z = [X, θ m Combine the data to construct a multiphase field model simulation dataset, and then normalize it to divide it into training and test sets;

[0052] (2) Set the number of rules for the fuzzy neural network and initialize the rule centers and widths. Then, train the fuzzy neural network using the training set until the error of the fuzzy neural network's prediction results on the test set meets the requirements. The error function of the fuzzy neural network is the root mean square error function (RMSE), and its calculation formula is as follows:

[0053]

[0054] Where n represents the number of test samples, Y i For the actual output result, The output result is predicted by the fuzzy neural network;

[0055] (3) For the latest obtained feed parameter vector X new and output parameter vector Y new The initial parameters of the NSGA-II algorithm are set, including population size, number of iterations, crossover and mutation probabilities. A trained fuzzy neural network is then used to perform multi-objective optimization on the coefficient multiplier θ to obtain the Pareto front solution set. The optimization objective is to minimize the iron content w in the bottom charge. Fe,out Temperature value T at the furnace wall temperature sampling point g,wall The error between the predicted and true values ​​of the fuzzy neural network is calculated; the best-in-class solution distance method (TOPSIS) is used to filter the Pareto front solution set to obtain the optimal solution θ of the coefficient multipliers. opt .

[0056] V. The optimal solution of the coefficient multipliers and the key vertical shaft furnace feed parameters are re-inputted into the multiphase field model to generate the optimized multiphase field simulation results of the hydrogen-based vertical shaft furnace:

[0057] Multiply the coefficients by the optimal solution θ opt and the corresponding key vertical shaft furnace feed parameter X new After combining the components, the multiphase field model constructed in step one is re-entered for solution to obtain the optimized multiphase field simulation results of the hydrogen-based vertical shaft furnace.

[0058] The beneficial effects of this invention are as follows:

[0059] This invention proposes a mechanism- and data-driven multiphase field modeling method for hydrogen-based shaft furnaces to improve the modeling accuracy of the ironmaking process. First, a preliminary physical model of the internal multiphase field of the hydrogen-based shaft furnace is constructed using fluid dynamics equations, providing a mechanistic constraint basis for the model. Second, a reaction model is constructed based on the additive reaction time law, and a coefficient multiplier is assigned to each characteristic time term, making it possible to use data to compensate for discrepancies between mechanistic assumptions and reality. Subsequently, the model is fully stimulated by randomly selecting values ​​for the coefficient multipliers to obtain simulation data for the multiphase field model. A surrogate model is constructed and the coefficient multipliers are optimized using a fuzzy neural network combined with a non-dominated sorting genetic algorithm II (NSGA-II) to obtain optimized simulation results for the multiphase field of the hydrogen-based shaft furnace. The proposed multiphase field modeling method for hydrogen-based shaft furnaces can achieve accurate modeling of the internal multiphase field of the furnace and has high simulation accuracy for different smelting conditions. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the one-dimensional multiphase field modeling route for a hydrogen-based vertical shaft furnace based on fluid mechanics and fuzzy neural network parameter optimization in an embodiment of the present invention. Detailed Implementation

[0061] To more clearly describe the technical content of the present invention, the following description, in conjunction with the accompanying drawings, further illustrates the specific implementation methods of the present invention.

[0062] like Figure 1 The diagram shown illustrates the one-dimensional multiphase field modeling approach for a hydrogen-based vertical shaft furnace based on fluid mechanics and fuzzy neural network parameter optimization according to the present invention. The method includes the following steps:

[0063] (1) Constructing a multiphase field model inside a hydrogen-based vertical shaft furnace can provide a deeper understanding of its complex physicochemical processes, including gas diffusion, convection transport, reduction reaction and particle flow, thereby optimizing the operation of the vertical shaft furnace and improving the system's operating efficiency and environmental benefits.

[0064] The momentum conservation equation is used to describe the gas pressure distribution inside the shaft furnace, mainly including the Ergun equation, which is expressed in the following form:

[0065]

[0066] Where the subscript g represents the gas phase; u g d represents the apparent gas flow rate, p represents the gas pressure, ε represents the porosity of the particle bed, and d represents the apparent gas flow rate. p Represents the diameter of ore particles, μ g and ρ g denoted by and , respectively, the viscosity and density of the gas phase, and z represents the axial direction of the vertical furnace.

[0067] Since the solid bed inside an actual hydrogen-based shaft furnace moves downwards slowly and almost no movement occurs in the cross-section, the solid velocity can be considered a fixed value, unaffected by factors such as pressure, and is expressed as:

[0068]

[0069] Where the subscript 's' represents the solid phase; W represents the mass flow rate of the solid; and ρ b S represents the apparent mass density of the bed, and S represents the cross-sectional area of ​​the vertical furnace.

[0070] The concentration distribution of gaseous and solid components inside the hydrogen-based vertical shaft furnace is described using a set of mass conservation equations, including the gas mass conservation equation, the reaction reduction degree distribution equation, and the relationship between the solid mass fraction and the reaction reduction degree. The gas mass conservation equation is expressed in the following form:

[0071]

[0072] The equation for the distribution of the degree of reduction of the reaction is expressed in the following form:

[0073]

[0074] In this context, the subscripts i and n represent the gas phase component and reaction number, respectively. i = H2 or H2O, and n = 1 to 3, representing the reactions 3Fe2O3 + H2 → 2Fe3O4 + H2O, Fe3O4 + H2 → 3FeO + H2O, and FeO + H2 → Fe + H2O, respectively; x i and X n R represents the mole fraction of gaseous component i and the degree of reaction of the nth reaction, respectively; n Represents the reaction rate of the nth reaction; n g Represents the gas phase molar flow rate; κ is the initial reducible oxygen content in a single particle; β is the initial oxygen molar fraction for each reduction step. n The values ​​are 0.1111, 0.1889, and 0.70, respectively.

[0075] Knowing the degree of reduction, the mass fraction of each solid component in the ore can be determined using the relationship between the solid mass fraction and the degree of reduction. This relationship is expressed as follows:

[0076]

[0077]

[0078] in, w FeO w Fe The values ​​represent the mass fractions of the solid phase components Fe2O3, Fe3O4, FeO, and Fe, respectively.

[0079] The temperature distribution of the gas and solid phases inside the hydrogen-based vertical shaft furnace is described using a set of energy conservation equations, including gas energy conservation equations and solid energy conservation equations. The gas energy conservation equation is expressed in the following form:

[0080]

[0081] The solid energy conservation equation is expressed in the following form:

[0082]

[0083] Among them, T g and T s These represent the gas phase and solid phase temperatures, respectively; G represents the gas volumetric flow rate; C g and C s These represent the isobaric specific heat capacities of the gas phase and solid phase, respectively; h p and h w These represent the gas-solid heat transfer coefficient and the overall heat transfer coefficient between the gas and the outside environment through the water-cooled furnace wall, respectively; D represents the internal diameter of the vertical furnace; T a Represents ambient temperature; ΔH n This represents the enthalpy change of the nth reaction.

[0084] (2) The construction of the reaction model based on the additive reaction time law includes the following steps:

[0085] The reaction model comprises three reactions: 3Fe₂O₃ + H₂ → 2Fe₃O₄ + H₂O, Fe₃O₄ + H₂ → 3FeO + H₂O, and FeO + H₂ → Fe + H₂O. Each reaction is divided into mass transfer stages and an analytical model is constructed using the additive reaction time law. The additive reaction time law states that when different transport resistances (e.g., chemical reactions, diffusion, external mass transfer, etc.) occur in series, the total reaction time required to complete the complete conversion of the solid reactants can be approximated as the sum of the characteristic times of each individual step. The characteristic time of each individual step is the time required to achieve complete conversion in a system controlled only by that basic step. In the hydrogen-based shaft furnace reaction model, these steps are: external mass transfer of gas through the boundary layer surrounding the particles, diffusion of gas through the pores of the particles, diffusion of gas within the grains of the particles, diffusion of gas within the crystals of the grains, and gas-solidification chemical reaction. The characteristic time of external mass transfer of gas through the boundary layer surrounding the particles is expressed as:

[0086]

[0087] The characteristic time of gas diffusion through particle pores is described as follows:

[0088]

[0089] The characteristic time of gas diffusion within the grains of a particle is described as follows:

[0090]

[0091] The characteristic time of gas diffusion within the crystal grain is described as follows:

[0092]

[0093] The characteristic time of gas-solidification chemical reactions is described as follows:

[0094]

[0095] Based on the characteristic time of each transmission resistance step, the formula for calculating the reaction rate is expressed as follows:

[0096]

[0097] Among them, c n k represents the concentration of the solid reactant in the nth reaction. ext Indicates the external mass transfer coefficient of the gas; These represent the effective diffusion coefficients of H2 within the particle, grain, and crystal, respectively; k ch,n and K n Let d represent the reaction rate constant and the reaction equilibrium constant of the nth reaction, respectively; gr and d cr These represent the diameters of the FeO grains and the crystals, respectively. and These represent the mole fractions of H2 and H2O outside the ore particles, respectively. This indicates the mole fraction of H2 at reaction equilibrium.

[0098] Based on this, a coefficient multiplier is assigned to each characteristic time term in the reaction model, as shown in the following equation:

[0099]

[0100] Where a, b, c, d, and e represent external mass transfer of gas through the boundary layer surrounding the particle, diffusion of gas through the pores of the particle, diffusion of gas within the grains of the particle, diffusion of gas within the crystals of the grains, and gas-solidification chemical reaction, respectively. The coefficient multiplier corresponding to the characteristic time term of external mass transfer of gas through the boundary layer surrounding the particle is used to adjust the weights of each term and eliminate the influence caused by deviations between the model parameter settings and the actual situation.

[0101] (3) The extraction of key variables of the vertical furnace and their combination with randomly sampled coefficient multipliers into the multiphase field model for simulation includes the following steps:

[0102] First, the feed and discharge parameters and key process variables of the vertical shaft furnace were extracted. The feed parameters of the vertical shaft furnace include the inlet gas velocity u of the H2 inlet pipe on the side of the vertical shaft furnace. g,in Intake pressure p in Intake air temperature T g,in The ore feed rate W at the top of the vertical shaft furnace and the feed temperature T s,in , Mass fraction of feed components w j,in Bed porosity ε and apparent mass density ρ b Particle diameter d p The discharge parameters of the vertical shaft furnace include the gas temperature T at the top of the furnace. g,out Pressure p out Component x i,out Iron content w at the bottom of the furnace Fe,out Discharge temperature T s,out The key process variables include the gas temperature T at the furnace wall temperature sampling point. g,wall The subscripts in, out, and wall represent the values ​​of the variable at the corresponding feed, discharge, and furnace wall sampling points, respectively.

[0103] After extracting the key parameters, the vertical shaft furnace feed parameters are combined with randomly sampled coefficient multipliers and input into the constructed multiphase field model. Specifically, the feed parameter vector X = [u g,in ,p in ,T g,in ,W,T s,in ,w j,in ,ε,ρ b ,d p ] and the coefficient multipliers {θ1,θ2,…,θ} of k randomly sampled groups m ,…,θ k}(where θ) m =[a m ,b m ,c m ,d m ,e m The parameters m∈[1,k], k≥5) are combined to obtain the model input vector group {[X,θ1],[X,θ2],…,[X,θ2],[X,θ3],[X,θ4],[X,θ5],[X,θ6],[X,θ7],[X,θ8],[X,θ9],[X,θ1 ...2],[X,θ1],[X, k Then, input vectors [X, θ] to each model are used. m Input the constructed multiphase field model into each of the above methods and solve them to obtain the corresponding multiphase field model solution results.

[0104] The randomly sampled coefficient multipliers are specifically coefficient multiplier values ​​randomly sampled within a predetermined range [0.1, 10]. They are used to artificially correct for ore particles, grains, crystal diameters, and mass transfer and diffusion coefficients that are difficult to determine precisely in the reaction model. For different feed parameter vectors, the coefficient multiplier group needs to be re-randomized to ensure that all values ​​of the coefficient multipliers cover the range as much as possible, achieving a sufficient excitation effect.

[0105] (4) The optimization of coefficient multipliers based on multiphase field model simulation data using fuzzy neural networks and NSGA-II algorithm includes the following steps:

[0106] First, the output parameter vector Y = [T] g,out ,p out ,x i,out ,w Fe,out ,T s,out ,T g,wall ] and the corresponding model input vector Z = [X, θ m Combine the data to construct a multiphase field model simulation dataset, and then normalize it to divide it into training and test sets;

[0107] Secondly, the number of rules in the fuzzy neural network is set, and the rule centers and widths are initialized. Then, the fuzzy neural network is trained using the training set until the error of its predictions on the test set meets the requirements. The error function of the fuzzy neural network is the root mean square error function (RMSE), and its calculation formula is as follows:

[0108]

[0109] Where n represents the number of test samples, Y i For the actual output result, The output result is predicted by the fuzzy neural network;

[0110] Subsequently, for the newly acquired feed parameter vector X new and output parameter vector Y new The initial parameters of the NSGA-II algorithm are set, including population size, number of iterations, crossover and mutation probabilities. A trained fuzzy neural network is then used to perform multi-objective optimization on the coefficient multiplier θ to obtain the Pareto front solution set. The optimization objective is to minimize the iron content w in the bottom charge. Fe,out Temperature value T at the furnace wall temperature sampling point g,wall The error between the predicted value and the true value of the fuzzy neural network;

[0111] Finally, the best-inferior-best-inferior solution distance method (TOPSIS) is used to filter the Pareto front solution set to obtain the optimal solution θ for the coefficient multipliers. opt .

[0112] (5) The step of re-inputting the optimal solution of the coefficient multiplier and the key vertical shaft furnace feed parameters into the multiphase field model to generate optimized multiphase field simulation results for the hydrogen-based vertical shaft furnace is specifically as follows: the optimal solution of the coefficient multiplier θ opt and the corresponding key vertical shaft furnace feed parameter X new After combining the components, the pre-constructed multiphase field model is re-inputted for solution, resulting in optimized multiphase field simulation results for the hydrogen-based vertical shaft furnace.

[0113] This invention proposes a mechanism- and data-driven multiphase field modeling method for hydrogen-based shaft furnaces to improve the modeling accuracy of the ironmaking process. First, a preliminary physical model of the internal multiphase field of the hydrogen-based shaft furnace is constructed using fluid dynamics equations, providing a mechanistic constraint basis for the model. Second, a reaction model is constructed based on the additive reaction time law, and a coefficient multiplier is assigned to each characteristic time term, making it possible to use data to compensate for discrepancies between mechanistic assumptions and reality. Subsequently, the model is fully stimulated by randomly selecting values ​​for the coefficient multipliers to obtain simulation data for the multiphase field model. A surrogate model is constructed and the coefficient multipliers are optimized using a fuzzy neural network combined with a non-dominated sorting genetic algorithm II (NSGA-II) to obtain optimized simulation results for the multiphase field of the hydrogen-based shaft furnace. The proposed multiphase field modeling method for hydrogen-based shaft furnaces can achieve accurate modeling of the internal multiphase field of the furnace and has high simulation accuracy for different smelting conditions.

[0114] The above description is not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A hydrogen-based shaft furnace one-dimensional multiphase field modeling method based on fuzzy neural network optimization, characterized by the steps of The application relates to a method for optimizing hydrogen-based shaft furnace process parameters, comprising the following steps: 1) constructing a one-dimensional multiphase field model describing the pressure, temperature and component concentration distribution of gas and solid phases in a hydrogen-based shaft furnace based on a fluid mechanics equation group; The fluid mechanics equation group comprises: a momentum conservation equation, an Ergun equation for describing the pressure drop of a gas phase; a gas mass conservation equation and a solid mass conservation / reduction degree distribution equation; and an energy conservation equation, comprising a gas energy equation and a solid energy equation; 2) constructing a reaction model based on an additive reaction time law, and assigning a coefficient multiplier to each characteristic time term in the model, and taking the improved reaction model as a source term in the multiphase field model; 3) extracting a plurality of groups of shaft furnace feeding and discharging parameters and key process variables under real industrial conditions, combining the shaft furnace feeding parameters with randomly sampled coefficient multipliers, and inputting the combination into the constructed multiphase field model to obtain corresponding multiphase field model solving results; 4) training a fuzzy neural network using the multiphase field model simulation data, and performing multi-objective optimization on the trained fuzzy neural network by combining a non-dominated sorting genetic algorithm II (NSGA-II), so as to obtain an optimal solution of the coefficient multiplier; 5) inputting the optimal solution of the coefficient multiplier and the key shaft furnace feeding parameters into the multiphase field model again to generate an optimized hydrogen-based shaft furnace multiphase field simulation result; All the above steps are executed by a processor of at least one general-purpose computer through a program, and real-time industrial field measurement data can be acquired for error evaluation.

2. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, characterized in that, In step 1), the equation for describing the gas phase pressure distribution in the hydrogen-based shaft furnace in the fluid mechanics equation group is a momentum conservation equation, which is described by an Ergun equation and is expressed in the following form: where the subscript g represents the gas phase; u g represents the superficial gas velocity, p represents the gas pressure, ε represents the porosity of the particle bed, d p represents the diameter of the ore particles, μ f and ρ g represent the viscosity and density of the gas phase, respectively, and z represents the vertical shaft direction; The solid phase velocity is expressed as: where the subscript s represents the solid phase; W represents the mass flow rate of the solids, p b represents the superficial mass density of the bed, and S represents the cross-sectional area of the shaft.

3. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, characterized in that, In step 1), the equation group for describing the gas and solid phase component concentration distribution in the hydrogen-based shaft furnace in the fluid mechanics equation group is a mass conservation equation group, comprising a gas mass conservation equation, a reaction reduction degree distribution equation and a relationship between the solid mass fraction and the reaction reduction degree, wherein the gas mass conservation equation is expressed in the following form: The reaction reduction degree distribution equation is expressed in the following form: where the subscripts i and n represent the gas phase component and the reaction order, respectively, i = H2or H2O, n = 1-3, representing the reactions 3Fe2O3+ H2→ 2Fe3O4+ H2O, Fe3O4+ H2→ 3FeO + H2O, FeO + H2→ Fe + H2O, respectively; x i and X n represent the mole fraction of the gas phase component i and the reaction extent of the nth reaction, respectively; r n represents the reaction rate of the nth reaction; n g represents the gas phase molar flow rate; κ is the initial reducible oxygen content in a single particle; the initial oxygen mole fraction β n are 0.1111, 0.1889, and 0.70, respectively. After the reaction reduction degree is known, the mass fraction of each solid phase component in the ore is obtained through the relationship between the solid mass fraction and the reaction reduction degree, and the relationship between the solid mass fraction and the reaction reduction degree is expressed as wherein, w FeO , w Fe are the mass fractions of the solid components Fe2O3, Fe3O4, FeO, Fe, respectively.

4. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, characterized in that, In step 1), the equation group for describing the gas and solid phase temperature distribution in the hydrogen-based shaft furnace in the fluid mechanics equation group is an energy conservation equation group, comprising a gas energy conservation equation and a solid energy conservation equation, wherein the gas energy conservation equation is expressed in the following form: The solid energy conservation equation is expressed in the following form: where T g and T s represent the gas and solid phase temperatures, respectively; G represents the gas volumetric flow rate; C g and C s represent the gas and solid phase constant pressure specific heat capacities, respectively; h p and h w represent the gas-solid heat transfer coefficient and the overall heat transfer coefficient between the gas and the environment through the water-cooled walls, respectively; D represents the internal diameter of the shaft furnace; T a represents the ambient temperature; ΔH n represents the reaction enthalpy of the nth reaction.

5. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, characterized in that, In step 2), the reaction model comprises three reactions, including 3Fe2O3+H2→2Fe3O4+H2O, Fe3O4+H2→3FeO+H2O, FeO+H2→Fe+H2O; each reaction is divided into mass transfer stages and a model is constructed by using the additive reaction time law, which states that when different transfer resistances occur in series, the total reaction time required for complete conversion of solid reactants can be approximately the sum of the characteristic times of each individual step, each of which is the time required to achieve complete conversion in a system controlled only by that elementary step; in the hydrogen-based shaft furnace reaction model, these steps are: external mass transfer of gas through the boundary layer around the particle, diffusion of gas through the particle porosity, diffusion of gas within the grain inside the particle, diffusion of gas within the crystal inside the grain, and gas-solid chemical reaction; the characteristic time of external mass transfer of gas through the boundary layer around the particle is expressed as: the characteristic time of diffusion of gas through the particle porosity is expressed as: the characteristic time of diffusion of gas within the grain inside the particle is expressed as: the characteristic time of diffusion of gas within the crystal inside the grain is expressed as: the characteristic time of gas-solid chemical reaction is expressed as: Based on the characteristic times of each transfer resistance step, the calculation formula of the reaction rate is expressed as where c n represents the concentration of the solid reactant for the nth reaction; k ext represents the external mass transfer coefficient of the gas; represents the effective diffusion coefficient of H2 inside the particle, inside the grain, and inside the crystal, respectively; k ch,n and K n represent the reaction rate constant and the reaction equilibrium constant for the nth reaction, respectively; d gr and d cr represent the diameter of the grain and the crystal inside FeO, respectively; and represent the mole fraction of H2 and H2O outside the ore particle, respectively; represents the mole fraction of H2 at the reaction equilibrium.

6. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, wherein, In step 2), a coefficient multiplier is given to each characteristic time term in the reaction model, as shown in the following formula: where a, b, c, d, and e represent the characteristic time terms of external mass transfer of gas through the boundary layer around the particle, diffusion of gas through the particle porosity, diffusion of gas within the grain inside the particle, diffusion of gas within the crystal inside the grain, and gas-solid chemical reaction, respectively; the coefficient multiplier corresponding to the characteristic time term of external mass transfer of gas through the boundary layer around the particle is used to adjust the weight of each term and eliminate the impact of deviations between model parameter settings and actual conditions.

7. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, characterized in that, In step 3), the shaft furnace input parameters include the shaft furnace side gas inlet pipe H2inlet flow rate u g,in , inlet pressure p in , inlet temperature T g,in , shaft furnace top ore feed rate W, feed temperature T s,in , feed composition mass fraction w j,in , bed porosity ε and apparent mass density ρ b , pellet particle diameter d p ; the shaft furnace output parameters include the shaft furnace top gas temperature T g,out , pressure p out , composition x i,out , furnace bottom discharge iron content w Fe,out , discharge temperature T s,out ; the key process variables include the gas temperature value T g,wall at the furnace wall temperature sampling point; the subscripts in, out and wall respectively represent the values of the variables at the corresponding input, output positions and furnace wall sampling point.

8. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, characterized in that, In step 3), the shaft furnace feeding parameters are combined with the randomly sampled coefficient multipliers and input into the constructed multiphase field model as follows: a feeding parameter vector X composed of each set of feeding parameters is combined with a set of k randomly sampled coefficient multipliers {θ1, θ2, …, θk}, where θm=[a, b, c, d, e], m∈[1, k], k≥5, to obtain a set of model input vectors {[X, θ1], [X, θ1], …, [X, θk]} corresponding to the set of feeding parameters, and then each model input vector [X, θm] is input into the multiphase field model constructed in step 1) for solving to obtain the corresponding multiphase field model solving result. g,in in g,in s,in j,in b p m k m m m m m m k m ​​​​​​​​​​​​​​​​​ The coefficient multiplier obtained by random sampling is a value randomly sampled within the predetermined value range [0.1, 10], which is used to artificially correct the diameters of ore particles, grains, and crystals, as well as mass transfer and diffusion coefficients that are difficult to accurately determine in the reaction model; for different feed parameter vectors, the coefficient multiplier group needs to be randomly sampled again, so that all values of the coefficient multiplier cover the value range as much as possible to achieve full stimulation.

9. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, wherein, In step 4), the optimization of the coefficient multiplier based on the multi-phase field model simulation data using fuzzy neural networks and the NSGA-II algorithm includes the following steps: (1) The output parameter vector Y = [T g,out ,p out ,x i,out ,w Fe,out ,T s,out ,T g,wall ] is combined with the corresponding model input vector Z = [X, θ m ] to construct a multi-phase field model simulation dataset, and the dataset is normalized and divided into a training set and a test set; (2) Set the number of rules of the fuzzy neural network and initialize the rule center and width, then train the fuzzy neural network using the training set until the error of the prediction results of the fuzzy neural network on the test set meets the requirements; the error function of the fuzzy neural network is the root mean square error function (RMSE), which is expressed as: wherein n represents the number of test samples, Y i is the true output result, is the fuzzy neural network predicted output result; (3) For the latest acquisition of the feed parameter vector X new and output parameter vector Y new , the initial parameters of the NSGA-II algorithm are set, including the population size, the number of iterations, the crossover and mutation probability, and the coefficient multiplier θ is optimized by combining the trained fuzzy neural network to obtain the Pareto front solution set, and the optimization goal is to minimize the error between the fuzzy neural network predicted value and the real value of the furnace bottom discharge iron content w Fe,out and the temperature value T g,wall at the furnace wall temperature sampling point; the Pareto front solution set is screened by the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to obtain the optimal solution θ opt of the coefficient multiplier.

10. The fuzzy neural network optimization based one-dimensional multiphase field modeling method of hydrogen-based shaft furnace according to claim 1, characterized in that, The step 5) is specifically: multiplying the coefficient multiplier optimal solution θ opt and the corresponding key shaft furnace feeding parameter X new After combination, re-input the multiphase field model constructed in step 1) to solve, and obtain the optimized hydrogen-based shaft furnace multiphase field simulation result.

Images