Pure hydrogen shaft furnace iron oxide reduction state prediction method based on gas-solid coupling reaction

Abstract

The invention discloses a method for predicting the reduction state of iron oxide in a pure hydrogen shaft furnace based on a gas-solid coupling reaction. The method comprises the following steps: firstly, describing a reduction reaction mechanism of pellets in the pure hydrogen shaft furnace through an unreacted nuclear model, and obtaining a gas-solid reaction rate; then, constructing a gas-solid coupling reaction model based on computational fluid dynamics, and solving a reaction process and a reduction region state; and finally, storing grid point data under different parameter conditions as a shaft furnace reduction reaction data set, establishing a data-driven space-time prediction model, and predicting the total iron content of the pellets by using process variables. According to the method, the numerical simulation calculation efficiency is improved, the experimental data acquisition process is simplified, and the reduction states of the iron oxide in the shaft furnace at different time and positions can be accurately predicted.

Description

Technical Field

[0001] The present invention belongs to the field of modeling and simulation of the hydrogen metallurgy production process, and particularly relates to a method for predicting the reduction state of iron oxides in a pure hydrogen shaft furnace based on gas-solid coupling reaction. Background Art

[0003] The hydrogen metallurgy system includes subsystems such as raw material gas preparation cycle, shaft furnace reduction, and cooling and discharging. It is a typical complex production system with multiphase flow coupling. Among them, the shaft furnace reduction process is the core part of the hydrogen metallurgy system. To achieve intelligent regulation and monitoring of the hydrogen metallurgy production process and ensure the safe and stable operation of the system, it is crucial to accurately characterize the reduction reaction state inside the pure hydrogen shaft furnace. However, the internal environment of the shaft furnace presents a "black box" state of being airtight, high-temperature, high-pressure, and multi-dust, making it difficult to characterize the coupling mechanism of multiphase flow, transfer, and reaction.

[0004] In the prior art, experimental means and numerical simulation are two main methods for studying the complex reduction process inside the pure hydrogen shaft furnace. Experimental means usually require building an experimental platform similar to the real production environment for analysis. However, it has high costs, a long cycle, is greatly affected by the accuracy of testing equipment and the environment, has limited data acquisition ability, and is difficult to comprehensively reflect the dynamic evolution law inside the shaft furnace.

[0005] The numerical simulation method mainly uses the computational fluid dynamics (CFD) method to provide the spatio-temporal distribution of physical parameters, divides the mesh of the shaft furnace reduction domain by means of finite element software, and constructs a differential equation group through a mechanism model for numerical iterative solution. Although this method can accurately simulate the actual situation of the reaction, it requires building a complex mathematical model and performing refined mesh division, resulting in an excessive amount of calculation and limiting the real-time performance of industrial applications.

[0006] With the accumulation of industrial data, data-driven modeling methods such as machine learning and neural network algorithms have received attention. They can learn the relationship between the operation parameters and reaction state of the shaft furnace from a large amount of data. Compared with the traditional numerical simulation method, they can evaluate and predict the reaction state of the shaft furnace more quickly. However, its effectiveness highly depends on the data quality, and it is necessary to ensure the authenticity and reliability of the shaft furnace operation parameters and reaction state data.

[0007] Currently, in the process of iron oxide reduction in a pure hydrogen shaft furnace, the acquisition of real working condition data is limited, and the existing experimental, simulation, and data-driven methods have limitations in terms of model accuracy, calculation efficiency, and data quality. Summary of the Invention

[0008] To overcome the deficiencies of existing modeling methods for the reduction process of iron oxides in a pure hydrogen shaft furnace, especially the low computational efficiency of numerical simulation methods and the lack of high-quality data in data-driven methods, the purpose of the present invention is to provide a method for predicting the reduction state of iron oxides in a pure hydrogen shaft furnace based on gas-solid coupling reactions, so as to achieve high-precision and rapid prediction of the reduction state inside the shaft furnace.

[0009] The technical solution for achieving the purpose of the present invention is as follows:

[0010] A method for predicting the reduction state of iron oxides in a pure hydrogen shaft furnace based on gas-solid coupling reactions, comprising the following steps:

[0011] 1) Use the unreacted core model to describe the reduction kinetics of the reaction between iron oxides and hydrogen, and solve the reaction rate;

[0012] 2) Establish a two-dimensional geometric model of the reaction domain in the shaft furnace and divide the grid. Set the iron oxides as a porous medium, and set the initial parameters and boundary conditions;

[0013] 3) Construct a computational fluid dynamics (CFD) gas-solid reaction numerical simulation model; based on the two-dimensional geometric model of the reaction domain in the shaft furnace in step 2), embed the reaction rate solved in step 1), and use the transient solution method for numerical iteration to obtain the component mass fraction at each time step of the grid points, and obtain the state of the reduction domain;

[0014] 4) Establish a spatio-temporal prediction model based on the numerical simulation data; save the grid point data at each time step under different parameter conditions in step 3) to form a shaft furnace reduction reaction data set, train a convolutional long short-term memory neural network (ConvLSTM) to extract the spatio-temporal features of the data, and achieve the prediction of the total iron content.

[0015] In step 1), the reduction kinetics is described by using a single-interface unreacted core model with step-by-step reactions, and then the reaction rate is solved, which is expressed as follows:

[0016]

[0017] Among them, R i is the reaction rate of the i-th chemical reaction, r s is the radius of the initial solid-phase pellet, is the initial concentration of hydrogen, is the hydrogen concentration after the reaction reaches the equilibrium state, k g is the mass transfer coefficient, k e is the diffusion coefficient, r i is the reaction interface radius of the i-th chemical reaction, K i is the reaction equilibrium constant of the i-th chemical reaction, k is the reaction rate constant, which is calculated using the Arrhenius formula, and the formula is as follows:

[0018]

[0019] Among them, A is the pre-exponential factor, E is the activation energy of the chemical reaction, R is the molar gas constant, and T is the reaction temperature.

[0020] In step 2), the establishment of the two-dimensional geometric model of the shaft furnace reaction zone and the mesh division are as follows:

[0021] Obtain the central section of the physical model of the reduction section of the pure hydrogen shaft furnace, establish a two-dimensional rectangular geometric model according to the actual parameters of the shaft furnace, and perform discretized mesh division on the geometric model to form m×n grid cells;

[0022] The iron oxide is set as a porous medium as follows:

[0023] The physical properties of the solid phase inside the shaft furnace are set as a porous medium material, and the relevant parameters include the porosity ∈ and the permeability ζ. The calculation formula is:

[0024]

[0025] ζ = cd 2 (4);

[0026] Among them, V p is the pore volume, V a is the total volume of the porous medium, c is the solid shape factor, and d is the average diameter of the solid particles;

[0027] The initial parameters and boundary conditions are set as follows:

[0028] The initial parameters and boundary conditions of the numerical simulation model include the input hydrogen concentration, the input hydrogen flow rate, the initial temperature of hydrogen, the initial temperature of iron ore pellets, the porosity of iron ore, the resistance coefficient, the internal pressure of the shaft furnace, and the temperature of the shaft furnace wall surface. The corresponding values are set according to the actual production conditions to represent different working conditions.

[0029] In step 3), the construction of the computational fluid dynamics (CFD) gas-solid reaction numerical simulation model is as follows:

[0030] The CFD numerical simulation model uses the laminar Euler-Euler model to describe the gas-solid two-phase flow in the shaft furnace. The basic equations include the mass conservation equation of mass transfer, the momentum conservation equation, and the energy conservation equation of heat transfer on the gas-solid surface; among them, the mass conservation equation of mass transfer is used to solve the change in the mass fraction of each component of the reaction, and is expressed in the following form:

[0031]

[0032] Among them, C i is the concentration of the i-th substance, j i is the diffusion rate of the i-th substance, u represents the flow velocity of the object, Ri is the reaction rate of the i-th substance, which can be obtained from the unreacted core model in step (1). is the Hamiltonian operator;

[0033] The momentum conservation equation is used to solve the flow velocity of each phase of the reaction and is expressed in the following form:

[0034]

[0035] where the subscript q represents the gas phase or the solid phase, and α q represents the phase volume fraction, and ρ q is the phase density, is the phase flow velocity, p is the static pressure, and F g,q represents the phase gravity, represents the external volume force;

[0036] The energy conservation equation for heat transfer at the gas-solid surface is used to calculate the temperature of the reaction domain and is expressed in the following form:

[0037]

[0038] where ρ is the mass density of the gas-solid mixture, C p is the specific heat capacity, λ is the thermal conductivity, and Q is the heat of reaction;

[0039] The established CFD model is imported into the Fluent software for numerical iterative solution. The transient solver is selected for solution, the solution algorithm is set to the pressure implicit operator splitting method (PISO), the discrete method of the basic equation uses the second-order difference format, the solution time step is set to t, and the solution accuracy is controlled within 10 -2 range.

[0040] In step 4), the spatio-temporal prediction model established based on the numerical simulation data is as follows:

[0041] During the solution process of the numerical simulation model of the gas-solid reaction in computational fluid dynamics (CFD), the component mass fraction data of the central axial grid points at all time steps are saved to form a time series data matrix, representing the reaction states in different spatio-temporal dimensions of the shaft furnace reduction domain; a convolutional long short-term memory neural network (ConvLSTM) spatio-temporal prediction model is constructed, including a convolutional layer, an LSTM layer, and a linear layer. The initial parameters and the changes in the component mass fraction of the grid points at each time step are used as spatio-temporal dimension information and input into the ConvLSTM to extract spatio-temporal features. The reduced total iron content in the future k time steps is used as the reduction state label to train the model; after the model training is completed, the initial parameters and the prediction time step are input to realize the prediction of the reduction state of iron oxides in the shaft furnace driven by data.

[0042] Advantageous effects of the present invention:

[0043] The present invention proposes a spatio-temporal modeling method for gas-solid coupling reactions in the hydrogen metallurgy production process, which is used for reduction state prediction. Specifically, first, aiming at the problem that it is difficult to characterize the coupling mechanism of multiphase flow, transfer, and reaction inside the shaft furnace, the shrinking core model is used to describe the reduction reaction mechanism, and high-quality simulation data is obtained through CFD numerical simulation methods, greatly reducing the cost of manual experiments and providing a reliable theoretical basis. Then, aiming at the problems of low computational efficiency and poor real-time performance of the numerical simulation method, a data-driven spatio-temporal modeling method is proposed. The CFD simulation grid point data set is saved to train the neural network model. The ConvLSTM model used has the ability of nonlinear fitting and time series processing, and can better capture the dynamic change law in the shaft furnace reduction process, while significantly improving the computational efficiency. The present invention realizes high-precision real-time prediction of the reduction state of iron oxides in the shaft furnace, which is helpful for the intelligent regulation and monitoring of the hydrogen metallurgy production process. Description of the Drawings

[0044] Figure 1 The flowchart of the present invention is shown as follows.

[0045] Figure 2 The schematic diagram of the shrinking core model is shown as follows.

[0046] Figure 3 The transient simulation results of the iron content in the two-dimensional reduction domain are shown as follows.

[0047] Figure 4 The structure diagram of the ConvLSTM spatio-temporal prediction model is shown as follows. Detailed Embodiments

[0048] The technical solution of the present invention will be further described below with reference to the drawings.

[0049] The purpose of the present invention is to provide a method for predicting the reduction state of iron oxides in a pure hydrogen shaft furnace based on gas-solid coupling reactions, which can accurately predict the reduction state characterized by the total iron content in real time according to the set process parameters during the hydrogen metallurgy process. The flowchart of the implementation method of the present invention is as Figure 1 shown below, and the steps are as follows:

[0050] (1) Use the shrinking core model to describe the reduction kinetics of the reaction between iron oxides and hydrogen, and solve the reaction rate;

[0051] (2) Establish a two-dimensional geometric model of the shaft furnace reaction domain and divide the grid, set the iron oxides as porous media, and set the initial parameters and boundary conditions;

[0052] (3) Construct a computational fluid dynamics (CFD) gas-solid reaction numerical simulation model; based on the shaft furnace geometric model in step (2), embed the reaction rate solved in step (1), and use the transient solution method for numerical iteration to obtain the component mass fraction at each time step of the grid points and obtain the state of the reduction domain;

[0053] (4) Establish a spatio-temporal prediction model based on numerical simulation data; save the grid point data of each time step under different parameter conditions in step (3) to form a shaft furnace reduction reaction dataset, and train a Convolutional Long Short-Term Memory Neural Network (ConvLSTM) to extract the spatio-temporal features of the data to achieve the prediction of the total iron content.

[0054] Step (1) describes the reduction kinetics based on the unreacted core model, solves the reaction rate, and provides a microscopic theoretical basis for CFD numerical simulation. Specifically, the iron ore pellets react with the countercurrent reducing gas in the shaft furnace, undergoing step-by-step reduction and finally becoming direct reduced iron. Among them, the following three main chemical reactions occur in the furnace:

[0055] 3Fe2O3 + H2 = 2Fe3O4 + H2O

[0056] Fe3O4 + H2 = 3FeO + H2O

[0057] FeO + H2 = Fe + H2O;

[0058] To describe the reduction kinetics of each stage of the reaction, the unreacted core model is used for analysis. The unreacted core model is as Figure 2 shown, which represents the interfacial chemical reaction between hydrogen and pellet ore. As the reaction progresses, the unreacted solid reaction core gradually shrinks. The gas-solid reaction rate can be calculated from the unreacted core model as follows:

[0059]

[0060] where, R i is the reaction rate of the i-th chemical reaction, r s is the radius of the initial solid-phase pellet, is the initial concentration of hydrogen, is the hydrogen concentration after the reaction reaches the equilibrium state, k g is the mass transfer coefficient, k e is the diffusion coefficient, r i is the reaction interface radius of the i-th chemical reaction, K i is the reaction equilibrium constant of the i-th chemical reaction, k is the reaction rate constant, which is calculated using the Arrhenius formula as follows:

[0061]

[0062] where, A is the pre-exponential factor, E is the activation energy of the chemical reaction, R is the molar gas constant, and T is the reaction temperature.

[0063] In step (2), first establish the geometric model and grid model of the shaft furnace, obtain the central cross-section of the physical model of the reduction section of the pure hydrogen shaft furnace, establish a two-dimensional rectangular geometric model according to the actual parameters of the shaft furnace, and perform discretized grid division on the geometric model to form m×n grid cells. The solid-phase physical properties inside the shaft furnace are set as porous medium materials. A porous medium refers to a material composed of solid substances that form a skeleton and have multiphase substances distributed in the internal pores, which conforms to the characteristics of pellet particles accumulating inside the shaft furnace and hydrogen being introduced. The relevant parameters include porosity ∈ and permeability ζ. Porosity represents the fluid transmission performance inside the porous medium, and permeability represents the ability of solid particles to allow fluid to pass through under a certain pressure difference. The calculation formulas are as follows:

[0064]

[0065] ζ = cd 2 ;

[0066] where, V p is the pore volume, V a is the total volume of the porous medium, c is the solid shape factor, and d is the average diameter of the solid particles.

[0067] After that, set the initial parameters and boundary conditions. The initial parameters and boundary conditions of the CFD numerical simulation model include the input hydrogen concentration, input hydrogen flow rate, initial hydrogen temperature, initial temperature of iron ore pellets, iron ore porosity, resistance coefficient, internal pressure of the shaft furnace, shaft furnace wall temperature, etc. Set the corresponding values according to the actual production conditions to represent different working conditions and provide initialization information for the CFD numerical solution.

[0068] Step (3) is used to construct a CFD gas-solid reaction numerical simulation model; the CFD numerical simulation model uses the laminar Euler-Euler model to describe the gas-solid phase flow inside the shaft furnace. The basic equations include the mass conservation equation of mass transfer, the momentum conservation equation, and the energy conservation equation of heat transfer at the gas-solid surface. Among them, the mass conservation equation of mass transfer is used to solve the change in the mass fraction of each component of the reaction and is expressed in the following form:

[0069]

[0070] where, C i is the concentration of the i-th substance, j i is the diffusion rate of the i-th substance, u represents the flow velocity of the object, R i is the reaction rate of the i-th substance, which can be obtained from the unreacted core model in step (1), is the Hamiltonian operator.

[0071] The momentum conservation equation is used to solve the flow velocity of each phase of the reaction and is expressed in the following form:

[0072]

[0073] Among them, the subscript q represents the gas phase or solid phase, and α q represents the phase volume fraction, ρ q is the phase density, is the phase flow velocity, p is the static pressure, and F g,q represents the phase gravity, represents the external volume force.

[0074] The energy conservation equation for heat transfer at the gas-solid surface is used to calculate the temperature of the reaction domain and is expressed in the following form:

[0075]

[0076] Among them, ρ is the mass density of the gas-solid mixture, C p is the specific heat capacity, λ is the thermal conductivity, and Q is the heat of reaction.

[0077] Based on the above equation, the established CFD model is imported into the Fluent software for numerical iterative solution. The transient solver is selected for the solution, the solution algorithm is set to PISO, the discrete method of the basic equation uses the second-order difference format, the solution time step is set to t, and the solution accuracy is controlled within 10 -2 range. With the numerical iterative solution, the mass fractions of each component in the reduction domain of the shaft furnace change and finally reach a steady state, providing high-quality data guarantee for the establishment of the data-driven model. The transient simulation results of the iron content at a certain time step are as Figure 3 shown.

[0078] In step (4), a spatio-temporal prediction model is established based on the numerical simulation data; during the solution process of the CFD numerical simulation model under different initial parameter conditions, the mass fraction data of the grid point components at the center axial direction of all time steps are saved to form a time series data matrix X, where X contains t time steps, p process variable dimensions, and m grid space dimensions, which is consistent with the grid division scale and represents the reaction states in different spatio-temporal dimensions of the reduction domain of the shaft furnace. The first 70% of the samples of X are set as the training set, and the last 30% of the samples are set as the test set for training the neural network model.

[0079] After that, a spatio-temporal prediction model of the convolutional long short-term memory neural network (ConvLSTM) is constructed, including a convolutional layer, an LSTM layer, and a linear layer. The network structure is as Figure 4 shown. The time series data matrix X including the initial parameters and the changes in the mass fractions of the grid point components at each time step is input into the ConvLSTM to extract spatio-temporal features. The convolutional operation uses a one-dimensional convolutional kernel to perform convolution on the grid space dimension, which is expressed as:

[0080] x′ t = Φ(x t * w + b);

[0081] Among them, xt Denote the input sequence of data at time step t, w is the convolution kernel weight, b is the bias term, and Φ is the activation function; the convolution operation can effectively extract the features of data in the spatial dimension and enhance the model's perception ability of local information.

[0082] After the convolution operation is completed, the data will flow into the LSTM layer. The LSTM layer, with its unique gating mechanism, can well handle the long-term dependencies in time series data, thereby further mining the time dimension features in the data. Its forward calculation formula is as follows:

[0083] f t = σ(W xf x t + W hf h t-1 + W cf c t-1 + b f )

[0084] i t = σ(W xi x t + W hi h t-1 + W ci c t-1 + b i )

[0085] c t = f t c t-1 + i t tanh(W xc x t + W hc h t-1 + b c )

[0086] o t = σ(W xo x t + W ho h t-1 + W co c t + b o )

[0087] h t = o t tanh(c t );

[0088] In the formula, σ is the activation function sigmoid; W is the variable weight coefficient, b represents the bias term; tanh is the hyperbolic tangent function; c t-1 is the memory cell state at time t-1; c t represents the updated memory cell state; ht is the output of the LSTM at time t, h t is input into the linear layer to obtain the predicted output y t .

[0089] During the training process, the reduced total iron content at the next k time steps is used as the reduction state label. Using these labels and the predicted output of the model, a mean squared error (MSE) loss function is constructed. Through the backpropagation algorithm, the network parameters are continuously adjusted with the goal of minimizing the loss function to gradually improve the prediction performance of the model.

[0090] In practical applications, the trained ConvLSTM model can receive new real-time data or simulated data and quickly output the corresponding predicted results of the reduced total iron content. These predicted results can provide timely and accurate guidance for process adjustment in the hydrogen metallurgy production process, improve production efficiency and product quality, and promote the intelligent development of the hydrogen metallurgy process.

[0091] The above embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but should not be construed as a limitation on the scope of the invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, which all belong to the protection scope of the present invention. The protection scope of the present invention is defined by the appended claims and any equivalent technical solutions thereof.

Claims

1. A method for predicting the reduction state of iron oxides in a pure hydrogen shaft furnace based on gas-solid coupling reaction, characterized in that: The following steps are involved: 1) Use the unreacted core model to describe the reduction kinetics of iron oxides and hydrogen and solve the reaction rate; 2) Establish a two-dimensional geometric model of the shaft furnace reaction domain and divide the grid, set the iron oxide as a porous medium, and set the initial parameters and boundary conditions; 3) Constructing a computational fluid dynamics (CFD) gas-solid reaction numerical simulation model; Based on the two-dimensional geometric model of the shaft furnace reaction domain in step 2), the reaction rate solved in step 1) is embedded, and a transient solution method is used for numerical iteration to obtain the mass fraction of the components at each time step of the grid point and obtain the state of the reduction domain; 4) Establishing a spatiotemporal prediction model based on numerical simulation data; saving the grid point data of each time step under different parameter conditions in step 3) to form a vertical furnace reduction reaction data set, training a convolutional long short-term memory neural network (ConvLSTM) to extract the spatiotemporal features of the data, and realizing the prediction of total iron content.

2. The method according to claim 1, characterized in that In step 1), the single interface unreacted core model of step-by-step reaction is used to describe the reduction kinetics, and then the reaction rate is solved, which is expressed as follows: Among them, R i is the reaction rate of the ith chemical reaction, r s is the radius of the initial solid phase pellet, is the initial concentration of hydrogen, is the hydrogen concentration after the reaction reaches equilibrium, k g is the mass transfer coefficient, k e is the diffusion coefficient, r i is the reaction interface radius of the ith chemical reaction, K i is the reaction equilibrium constant of the i-th chemical reaction, and k is the reaction rate constant, which is calculated using the Arrhenius formula as follows: Among them, A is the pre-exponential factor, E is the activation energy of the chemical reaction, R is the molar gas constant, and T is the reaction temperature.

3. The method according to claim 1, characterized in that In step 2), the two-dimensional geometric model of the shaft furnace reaction domain is established and the grid is divided as follows: Obtain the central section of the physical model of the pure hydrogen shaft furnace reduction section, establish a two-dimensional rectangular geometric model according to the actual parameters of the shaft furnace, and perform discretization grid division on the geometric model to form m×n grid units; The iron oxide is set as a porous medium as follows: The solid phase properties inside the shaft furnace are set as porous medium materials. The relevant parameters include porosity ∈ and permeability ζ, and the calculation formula is: ζ=cd 2 (4); Among them, V p is the pore volume, V a is the total volume of the porous medium, c is the solid shape factor, and d is the average diameter of the solid particles; The initial parameters and boundary conditions are set as follows: The initial parameters and boundary conditions of the numerical simulation model include input hydrogen concentration, input hydrogen flow rate, initial hydrogen temperature, initial temperature of iron ore pellets, iron ore porosity, resistance coefficient, vertical furnace internal pressure, and vertical furnace wall temperature. Corresponding values ​​are set according to actual production conditions to represent different operating conditions.

4. The method according to claim 1, characterized in that: In step 3), the construction of a computational fluid dynamics (CFD) gas-solid reaction numerical simulation model is as follows: The CFD numerical simulation model uses the laminar Euler-Euler model to describe the gas-solid phase flow in the shaft furnace. The basic equations include the mass conservation equation for material transfer, the momentum conservation equation, and the gas-solid surface heat transfer energy conservation equation. The mass conservation equation for material transfer is used to solve the changes in the mass fractions of the reaction components and is expressed as follows: Among them, C i is the concentration of the ith substance, j i is the diffusion rate of the ith substance, u represents the flow rate of the object, R i is the reaction rate of the ith substance, which can be obtained from the unreacted core model in step (1), is the Hamiltonian operator; The momentum conservation equation is used to solve the flow rate of each phase of the reaction and is expressed as follows: Wherein, the subscript q represents the gas phase or solid phase, α q represents the phase volume fraction, ρ q is the phase density, is the phase velocity, p is the static pressure, F g，q represents the phase gravity, represents the external body force; The energy conservation equation for heat transfer between gas and solid surfaces is used to calculate the temperature of the reaction domain and is expressed as follows: Where ρ is the material density of the gas-solid mixture, C p is the specific heat capacity, λ is the thermal conductivity, and Q is the reaction heat; The established CFD model was imported into Fluent software for numerical iterative solution calculation. The transient solver was selected for solution. The solution algorithm was set to pressure implicit operator separation (PISO). The basic equation discretization method used the second-order difference format. The solution time step was set to t, and the solution accuracy was controlled within 10 -2 within the range.

5. The method according to claim 1, characterized in that In step 4), the spatiotemporal prediction model is established based on numerical simulation data as follows: During the solution of the CFD gas-solid reaction numerical simulation model, the component mass fraction data of the central axial grid points of all time steps are saved to form a time series data matrix, which represents the reaction state of different spatiotemporal dimensions in the reduction domain of the shaft furnace. A convolutional long short-term memory neural network (ConvLSTM) spatiotemporal prediction model is constructed, including convolutional layer, LSTM layer and linear layer. The initial parameters and the changes in the mass fraction of the grid point components at each time step are used as spatiotemporal dimension information and input into ConvLSTM to extract spatiotemporal features. The reduced total iron content in the next k time steps is used as the reduction state label to train the model. After the model training is completed, the initial parameters and the predicted time step are input to realize the data-driven prediction of the reduction state of iron oxides in the shaft furnace.

Images