A Mechanism- and Data-Based Intelligent Modeling Method for Blast Furnaces

By combining mechanism and data in the intelligent modeling method of blast furnaces, the boundary conditions and calculation speed are optimized, solving the problems of accuracy and computational complexity in blast furnace temperature field modeling, and realizing accurate distribution and rapid calculation of blast furnace temperature field.

CN117786972BActive Publication Date: 2026-11-10NORTHEASTERN UNIV CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311805886.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-26
Publication Date
2026-11-10
Estimated Expiration
2043-12-26

AI Technical Summary

Technical Problem

Existing blast furnace temperature field modeling methods suffer from fuzzy boundary condition settings, complex and computationally intensive calculations, and an inability to accurately characterize the temperature field distribution in the furnace body. Furthermore, the accuracy of data-driven methods decreases when data is insufficient or conditions change.

Method used

A blast furnace intelligent modeling method combining mechanism and data is proposed. By establishing a temperature field distribution model of the inner surface of the furnace wall and a two-dimensional temperature field distribution model of the blast furnace body, an improved particle swarm optimization algorithm and a POD order reduction algorithm are adopted to optimize boundary conditions and calculation speed.

Benefits of technology

It improves the accuracy and calculation speed of blast furnace temperature field modeling, simplifies the calculation of tuyeres swirl zone, and enhances real-time performance and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117786972B_ABST
    Figure CN117786972B_ABST
Patent Text Reader

Abstract

This invention provides a mechanism- and data-based intelligent modeling method for blast furnaces, relating to the field of blast furnace smelting technology. The invention establishes a temperature field distribution model of the inner surface of the furnace wall to obtain the temperature field distribution of the inner surface of the furnace wall. Then, it establishes and solves a two-dimensional temperature field distribution model of the blast furnace body. Based on the physical and physicochemical parameters of the blast furnace to be tested, the thermocouple conditions inside the blast furnace wall are determined, resulting in the corresponding temperature distribution of the inner surface of the furnace wall and the two-dimensional temperature distribution of the furnace body.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of blast furnace smelting technology, and in particular to a method for intelligent modeling of blast furnaces based on mechanisms and data. Background Technology

[0002] The theoretical research on blast furnace temperature has a long history, mainly including expert system methods, physicochemical mechanism methods, mathematical methods, and methods combining mechanism and data. Expert system methods visualize expert experience, detailing operations and eliminating operational differences caused by different personnel, thus improving the stability of blast furnace operation. However, expert system methods rely on large amounts of data and knowledge; when the knowledge base is insufficient, the expert system cannot make correct judgments. Furthermore, the results of expert system methods are mostly representational and cannot provide an intuitive temperature field distribution. Mechanistic methods describe in detail the multiphase flow and interphase transfer within the blast furnace. However, the internal mechanisms of the blast furnace are complex and highly coupled. Therefore, general mechanistic methods make significant assumptions, such as treating density and heat transfer coefficient as constants and completely ignoring chemical reactions. Although the mechanism ensures the accuracy of temperature trends to some extent, the overall bias in the results may be large due to the large number of assumptions. Moreover, mechanistic models are mostly local models, analyzing only a part of the blast furnace and ignoring the overall blast furnace. Most blast furnace temperature estimation methods establish the relationship between input and output using data-driven approaches. Feature extraction and other methods approximate the quantitative relationship between the input and the furnace temperature or other variables that characterize it, and this relationship is used to predict and characterize the furnace temperature at the next time step. This method skips the analysis of the blast furnace mechanism, predicting future data based on existing historical data, thus avoiding the complex blast furnace mechanisms involved in mechanistic methods. If the established relationship is accurate enough, its output is significant for guiding blast furnace regulation and ensuring stable operation. However, its accuracy decreases linearly when the dataset size and data conditions are unsuitable. Even if the dataset is suitable, changes in production conditions can still affect accuracy. If the established relationship itself is sound and the production conditions are met, the result merely characterizes the furnace temperature using other data, failing to reveal the detailed distribution of the furnace temperature.

[0003] In recent years, with the development of computer hardware and algorithms, the combination of mathematics and physics has become a mainstream method for establishing the temperature field of blast furnaces. However, the current mainstream methods have the following drawbacks: the boundary conditions are relatively vague and have large errors; they cannot characterize the detailed temperature field distribution of the furnace body; most current methods use fluid simulation software such as Fluent to perform iterative calculations using the discrete method, which is complex, computationally intensive, slow, and has a large lag. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for intelligent modeling of blast furnaces based on mechanisms and data.

[0005] A mechanism- and data-based intelligent modeling method for blast furnaces includes the following steps:

[0006] Step 1: Establish a model of the temperature field distribution on the inner surface of the furnace wall;

[0007] Step 1.1: Obtain an overview of the furnace wall temperature;

[0008] Several thermocouples are installed at different heights on the blast furnace wall. The thermocouples at the same height are evenly distributed circumferentially between the cooling pipe and the inner wall of the blast furnace. The temperature at the corresponding position on the furnace wall is measured through the data feedback of the thermocouples, and the physical parameters of the furnace wall and the temperature parameters of the thermocouples inside the furnace wall are determined.

[0009] Step 1.2: Establish and discretize the differential equation of heat conduction of the blast furnace wall;

[0010] The differential equation for heat conduction is derived from the conservation of energy in solids and Fourier's law, which is:

[0011]

[0012] Where λ is the thermal conductivity, T is the temperature, and x, y, z are the coordinate directions of the blast furnace.

[0013] The heat conduction differential equation in the rectangular coordinate system in equation (1) is obtained by transforming x = rcosθ, y = rsinθ, and z = h:

[0014]

[0015] Where r is the radius of the circular cross-section of the blast furnace wall, θ is the angle of the circular cross-section of the blast furnace wall, and h is the height of the blast furnace.

[0016] Divide the blast furnace wall into several cross-sections to form several two-dimensional cross-sectional heat conduction models. For each two-dimensional cross-sectional heat conduction model, ignoring the longitudinal direction, we obtain:

[0017]

[0018] Where r0 is the radius of the inner ring of the two-dimensional cross-sectional heat conduction model, and r2 is the radius of the outer ring of the two-dimensional cross-sectional heat conduction model.

[0019] The partial differential equation is discretized into an algebraic equation using the finite difference method. Since the target region is regular, it is divided into a grid with equal step sizes, and the algebraic values ​​on the grid are used to replace the continuous numerical domain. The target domain is divided into M steps in the radial direction and N steps in the circumferential direction. There are N*(M+1) discrete points in the entire target domain, including the inner and outer boundaries. Then, equation (3) is discretized by central difference, and the discretized heat conduction differential equation is obtained as follows:

[0020]

[0021] Where i and j are the subscripts of the radial and circumferential discrete points, respectively, Δr is the radial step size of the two-dimensional cross-sectional heat conduction model, Δθ is the circumferential step size of the two-dimensional cross-sectional heat conduction model, and T... i,j Let r be the temperature of the grid point with index i,j after discretization. i Let i be the radius of the grid point with index i.

[0022] Step 1.3: Determine the boundary conditions of the blast furnace wall;

[0023] Including the inner surface boundary of the furnace wall:

[0024] T 1,j =T in,j , r=r0,0≤j≤N (5)

[0025] Boundary at the junction of refractory bricks and cooling wall:

[0026]

[0027] Discretized:

[0028]

[0029] Cooling wall outer surface boundary:

[0030]

[0031] Discretized:

[0032]

[0033] In the formula, T in This refers to the temperature distribution at the inner boundary, i.e., the temperature distribution on the inner surface of the blast furnace wall; T out This refers to the temperature distribution at the outer boundary, specifically the temperature distribution at the junction of the blast furnace cooling wall and the cooling pipes; T CW T represents the average temperature of the cooling water in the cooling pipes. CW =(T CW,in +T CW,out ) / 2, T CW,in T represents the temperature measured by the thermocouple at the cooling water inlet. CW,outTemperature measured by thermocouple at the cooling water outlet; h x v is the convective heat transfer coefficient between the cooling wall and the cooling water. 水 Cooling water flow rate:

[0034] h x =208 + 47.5v 水 (10)

[0035] r0, r1, and r2 are the physical parameters of the blast furnace wall. r0 is the inner diameter of the inner ring of the two-dimensional cross-sectional heat conduction model, which is the radius of the inner wall of the blast furnace. r1 is the radius at the junction of the refractory brick and the cooling wall. r2 is the radius of the outer ring of the two-dimensional cross-sectional heat conduction model, which is the radius at the location of the cooling pipe. λ1 is the thermal conductivity of the refractory brick, and λ2 is the thermal conductivity of the cooling wall.

[0036] Step 1.4: Use the optimized PSO algorithm to solve for the circumferential one-dimensional temperature distribution on the inner surface of the furnace wall;

[0037] The basic formula for the PSO algorithm is:

[0038]

[0039] In the formula, i is the i-th particle, i = 1, 2, ..., N; ω is the inertia weighting factor; V i (t) represents the velocity of the i-th particle in the t-th iteration; V i (t+1) is the velocity of the i-th particle in the (t+1)-th generation; X i (t) represents the position of the i-th particle in the t-th generation; P i Pg(t) represents the optimal position of a single particle in generation t; Pg(t) represents the global optimal position of the particle swarm in generation t; C1 and C2 are acceleration coefficients; R1 and R2 are random numbers uniformly distributed in [0,1].

[0040] The inertia weighting factor and acceleration coefficient in the PSO method have been redefined as follows:

[0041]

[0042] In the formula ω ini Let ω be the initial value. fin The final value of ω; M is an integer representing the number of iterations, C 1,ini and C 1,fin These represent the initial and final values ​​of C1, respectively; C 2,ini and C 2,fin Let ω represent the initial and final values ​​of C2, respectively. ini C 1,ini and C 2,fin For a larger value, ω fin C 1,finand C 2,ini It is a relatively small value.

[0043] The particle diversity(t) is defined as:

[0044]

[0045] In the formula: L max and L min Let represent the maximum and minimum values ​​of the search space, respectively; j represents the parameter in each particle; and i represents the particle number.

[0046] The search strategy for the next particle time step is selected based on the differences:

[0047] (1)diversity(t)>ξ||R3>a:

[0048]

[0049] (2)diversity(t)≤ξ&R3≤a:

[0050]

[0051] In the formula: R1, R2, R3, R4, and R5 represent random numbers uniformly distributed in [0,1]; ξ represents the constraint value of diversity(t); and a is a parameter that controls the initialization probability of particles.

[0052] Using the optimized PSO algorithm described above, the difference between the actual and calculated values ​​of the thermocouples inside the furnace wall is used as the objective function to solve the two-dimensional cross-sectional heat conduction model of the blast furnace wall composed of the heat conduction differential equation obtained in step 1.2 and the boundary conditions obtained in step 1.3, so as to obtain the one-dimensional circumferential temperature distribution of the inner surface of the multi-layer furnace wall.

[0053] Step 1.5: Solve for the one-dimensional longitudinal temperature distribution on the inner surface.

[0054] The blast furnace is divided into three parts: the throat and upper part of the furnace body, the lower part of the furnace body, and the waist and belly. Then, the temperature distribution of each part of the blast furnace in the longitudinal direction is fitted using a cubic curve calculation formula.

[0055] T w (θ j ,h)=β 0,j +β 1,j h+β 2,j h 2 +β 3,j h 3 , j=1,2,...,N (16)

[0056] Since the number of layers in each part is not consistent, in order to make full use of the data of each layer, Equation (17) is used for fitting to obtain the one-dimensional longitudinal temperature distribution at different angles on the inner surface of the blast furnace wall.

[0057]

[0058] In the formula: h represents the height of the blast furnace at different locations. i Let T be the height of the i-th layer. w (θ j h) represents the temperature at different locations within the blast furnace; different angles and heights correspond to different positions on the inner surface. β 0,j β 1,j β 2,j β 3,j represents the coefficients to be fitted to the cubic curve, and n represents the number of layers in the current part.

[0059] Thus, the one-dimensional longitudinal temperature distribution at all angles of the inner surface of the furnace wall is obtained. Combining these, the temperature field distribution of the inner surface of the furnace wall is obtained.

[0060] Step 2: Establish a two-dimensional temperature field distribution model of the blast furnace body;

[0061] Step 2.1: Use fluid theory and the Navier-Stokes equations to describe the blast furnace gas, analyze the flow, heat transfer and chemical reactions of the gas and solid phases, and establish a two-dimensional temperature field model of the blast furnace body by establishing partial differential equations describing four variables: pressure difference in the blast furnace body, gas velocity, gas temperature and solid temperature. The two-dimensional temperature field model of the blast furnace body is divided into three sub-models according to the coupling degree: a gas flow model describing gas pressure and velocity; a gas temperature model describing gas temperature; and a solid temperature model describing solid temperature.

[0062] (1) The gas flow model consists of a continuity equation and a momentum conservation equation that takes momentum transfer into account:

[0063]

[0064]

[0065]

[0066] In the formula: x and y represent the radial and axial directions, respectively; u and v represent the radial and axial velocities, respectively; p represents pressure; t represents time; and β represents the diffusion coefficient. μ is the kinetic viscosity coefficient, ρ is the gas density, and f1 and f2 are source term coefficients.

[0067]

[0068] In the formula η gε is the gas viscosity; s The porosity of the furnace charge; ψ s d is the shape factor of the solid particles; s The diameter of the solid particle;

[0069] (2) The gas temperature model is represented by the energy conservation equation, taking into account energy transfer and chemical reactions:

[0070] The gas temperature model considering energy transfer is as follows:

[0071]

[0072] In the formula: T g T represents the gas temperature. s Let S be the solid temperature, S be the effect of the source term on the gas temperature model, and α be the diffusion term coefficient. k is the thermal conductivity according to Fourier's law, c v γ is the specific heat, and γ is the heat transfer coefficient.

[0073]

[0074] In the formula: A g-s k is the contact area. g d is the thermal conductivity of the gas; s Pr is the average particle size; Re is the Reynolds number; Pr is the average particle size. g The Planck number represents the convective heat transfer number of the gas. Represents the energy effect of indirect reduction reactions on gases, unit: W / m³ 3 ; Represents the effect of direct reduction reaction on gases, unit: W / m³ 3 .

[0075] (3) The solid temperature model is represented by the heat conduction equation, taking into account gas-solid heat transfer and chemical reactions:

[0076]

[0077] In the formula: Represents the energy effect of indirect reduction reactions on solids, unit: W / m 3 ; The effect of direct reduction reaction on solids, unit: W / m 3 .

[0078] Step 2.2: Determine the modeling region and boundary conditions for the two-dimensional temperature field model of the blast furnace, where the boundary conditions include:

[0079] Inlet: The air vent is the inlet boundary of the model;

[0080] Outlet: The furnace top is set as a pressure outlet, and the parameters of this boundary, except for pressure, are obtained by interpolation from inside the modeling region;

[0081] Furnace wall: The flow at the wall surface is set to a no-slip boundary, and the furnace wall is impenetrable; therefore, the furnace wall velocity is set to 0. Heat transfer is set to convective heat transfer, i.e., the gas exchanges heat with the furnace wall via convection.

[0082] Symmetry axis: The blast furnace is a symmetrical model. The gradient of the vertical velocity at the symmetry axis is set to 0, and the magnitude of the normal velocity is 0.

[0083] Step 3: Solve the two-dimensional temperature field model of the blast furnace body;

[0084] Step 3.1: Mesh the modeling region of the two-dimensional temperature field distribution model of the blast furnace body;

[0085] The modeling region is located between the lower part of the tuyeres and the cross-shaped temperature measurement point in the furnace throat. An elliptic partial differential equation is used to generate the mesh for this region. Mesh convergence is determined by the maximum difference between the physical plane coordinates of the current iteration and the previous iteration, as shown in the following formula:

[0086]

[0087] Where e is the convergence error. This represents the maximum difference between the current round and the previous round's physical plane coordinates x-axis. This represents the maximum difference between the physical plane coordinates y-axis of the current round and the previous round.

[0088] The BFC mesh is generated by solving the Poisson equation. Let (x,y) be the physical plane coordinate system and (ξ,η) be the computational plane coordinate system. After transformation, the equations of the Poisson equations are solved using the computational plane to obtain the equations of the physical plane:

[0089]

[0090] In the formula: J, α, β, γ are coefficients:

[0091]

[0092] P(x,y) and Q(x,y) are source terms:

[0093]

[0094] The formulas for calculating φ and ψ on the boundary are as follows:

[0095]

[0096] After obtaining φ and ψ on the boundary, the values ​​of φ and ψ of the internal points are obtained by linear fitting. Then, P and Q of all grid points are obtained according to equation (28). At the same time, J, α, β and γ are calculated to obtain the relationship between the physical plane and the computational plane.

[0097] Step 3.2: Solve the two-dimensional temperature field of the blast furnace body using the SIMPLE algorithm;

[0098] The SIMPLE algorithm calculates the velocity based on the assumed pressure distribution using the momentum equation, and then uses the pressure correction equation for iterative correction so that the velocity satisfies both the momentum equation and the continuity equation.

[0099] First, the velocity before correction is obtained from the momentum equation, explained using radial velocity u; v is similarly derived from u. Discretizing the momentum equation, we have:

[0100]

[0101] In the formula The velocity term before calculation point correction; This is a coupling term for the velocities of adjacent points before correction; b e For source terms; This is the gradient term for the pressure correction value. Where a e Velocity before calculation point correction The coefficient, similarly, a nb and A e All of these are coefficients of their suffix parameters. The subscripts e / E, w / W, n / N, and s / S represent the left, right, lower, and upper boundaries of the main control unit grid, and p represents the current pressure point.

[0102] The corrected velocity and pressure satisfy the momentum equation, so:

[0103]

[0104] Where u e ′、u nb ′、P P ′ and P E ′ are the velocity and pressure correction values ​​for the calculation point and adjacent points, respectively.

[0105] Therefore:

[0106]

[0107] Considering only the gradient term of the pressure correction value, we have:

[0108] a e u e ′=A e (P P ′-P E′) (33)

[0109] Therefore, the velocity correction equation is:

[0110]

[0111] In the formula:

[0112]

[0113] At the same time, the corrected pressure and velocity need to satisfy the continuity equation, therefore:

[0114] (u e -u w )Δy+(v n -v s Δx=0 (36)

[0115] In the formula: Δx is the radial step size, and Δy is the axial step size.

[0116] Substituting equation (34) into equation (36), we get the pressure correction equation:

[0117] a P P P ′=a E P E ′+a W P W ′+a N P N ′+a S P S ′+b P (37)

[0118] In the formula: subscripts E, W, N, and S represent the neighboring points in the left, right, down, and up directions of the calculation point, and coefficient a P a E a W a N and a S for:

[0119]

[0120] b P It reflects the distance to convergence;

[0121]

[0122] Step 3.3: Use the POD order reduction algorithm to accelerate the solution speed of the two-dimensional temperature field model of the blast furnace body;

[0123] Step 3.3.1 Constructing the POD base

[0124] Assuming the two-dimensional temperature field model of the blast furnace body has a total time step count of T, N steps in the horizontal direction and M steps in the vertical direction, and a total number of nodes of m; firstly, the parameters of the SIMPLE algorithm solution process with L time steps are taken, including radial velocity u, axial velocity v, pressure p, and gas temperature T. g and solid temperature T s The distributions, 0 < L ≤ m, m = MN, respectively form the following five matrices A. u =(u i ) m*L A v =(v i ) m*L A p =(p i ) m*L A tg =(tg i ) m*L and A ts =(ts i ) m*L tg and ts represent the gas temperature and solid temperature variables, respectively.

[0125]

[0126]

[0127] Seeking and eigenvectors and eigenvalues;

[0128] The POD basis is constructed by taking several column vectors from the eigenmatrix formed by the eigenvectors.

[0129] Step 3.3.2: Define and solve for the current POD eigenvalues;

[0130] Assume the solution to the temperature model of the entire blast furnace body is: and

[0131] Then, they converge to

[0132]

[0133] Define the POD feature value as:

[0134]

[0135] The POD eigenvalues ​​for the previous L time period are calculated based on the current data and the corresponding POD basis.

[0136] Step 3.3.3: Solve for the POD eigenvalue at the next time step; based on the two-dimensional temperature field model of the blast furnace body, we get:

[0137]

[0138] in This represents the set of differences between the parameters of the current time step and the previous time steps, obtained from the discretized two-dimensional temperature field model of the blast furnace body.

[0139] According to equations (40) and (41), we obtain

[0140]

[0141] According to equations (42) and (43):

[0142]

[0143] in

[0144]

[0145] The POD eigenvalue at the next moment is obtained according to equation (44).

[0146] Step 3.3.4: Obtain the parameters for the next moment, including axial velocity, longitudinal velocity, pressure, gas temperature, and solid temperature distribution until convergence;

[0147] Equation (46) is derived from equation (41). Combining the POD eigenvalues ​​and the POD basis at the next time step, the parameter distribution at the next time step is obtained.

[0148]

[0149] Then continue solving until convergence to obtain the target value of the blast furnace internal temperature model.

[0150] Step 3.3.5: POD base update;

[0151] There is a POD error of

[0152]

[0153] have to

[0154]

[0155] According to equation (40), we have

[0156]

[0157] According to the definition of differential convergence, we have:

[0158] Under the premise of ensuring the stability of the discrete equations, according to equations (48) and (49), we get

[0159]

[0160] in and These represent the radial velocities at time steps n and n-1, radial step size i, and axial step size j, respectively. and These represent the radial velocities at time steps n and n-1, radial step size i, and axial step size j, respectively; Δt is the time step; Δx is the radial step; Δy is the axial step; and e is the axial step. n+1 and e n These represent the errors at time steps n+1 and n, respectively.

[0161] Under the condition of ensuring the stability of the equations, the following error estimate exists between the accurate solution of the two-dimensional temperature distribution model of the blast furnace and the numerical solution of the POD order reduction algorithm:

[0162]

[0163] in

[0164]

[0165] If satisfied

[0166]

[0167] Then δ≤1.

[0168] If you select M u M v M p M tg and M ts satisfy Then C n (δ)=(1+δ) n-L This is the decisive factor in determining POD base updates. Therefore, the following conclusion can be drawn:

[0169] if This necessitates using a new snapshot matrix to determine the new POD base instead of the old one.

[0170] Step 4: Determine the thermocouple conditions inside the blast furnace wall based on the physical and physicochemical parameters of the blast furnace to be tested, and obtain the temperature distribution on the inner surface of the furnace wall and the two-dimensional temperature distribution of the furnace body.

[0171] The beneficial effects of adopting the above technical solution are as follows:

[0172] (1) Mechanism and data analysis of the blast furnace wall were conducted. The heat transfer model, combined with thermocouples and an improved particle swarm optimization algorithm, was used to calculate the temperature of the inner wall of the blast furnace, thereby improving the accuracy of the model boundary condition settings and thus improving the accuracy of the model. The combustion reaction in the tuyere swirl zone is complex. An infrared camera was used to capture images from the tuyere mirror to measure the temperature of the tuyere swirl zone. This temperature was used to replace the blast temperature, making the inlet boundary conditions more accurate. At the same time, the influence of the combustion reaction in the tuyere swirl zone on the temperature could be ignored, simplifying the calculation and improving the accuracy.

[0173] (2) Adopting the single-phase flow simulation approach, the blast furnace is analyzed using single-phase flow, and the mass, momentum and energy transfer of the single-phase flow are analyzed. This approach can better meet the actual needs of the field with lower computational costs and improve real-time performance.

[0174] (3) In the process of iterative solution of traditional model, the POD algorithm is combined to reduce the order of the model, speed up the calculation speed of the model, reduce the calculation cost, and improve the real-time performance.

[0175] In general, this invention starts from the mechanism and combines images, data and mathematical methods to obtain a method for calculating the two-dimensional temperature field of the furnace body while improving the calculation speed under the premise of ensuring a certain calculation accuracy. Attached Figure Description

[0176] Figure 1 This is a schematic diagram of the discretization of the two-dimensional heat conduction model of the furnace wall cross section in an embodiment of the present invention;

[0177] Figure 2 This is a schematic diagram of furnace wall segmentation in an embodiment of the present invention;

[0178] Figure 3 This is a modeling area diagram of the blast furnace in an embodiment of the present invention;

[0179] Figure 4 This is a diagram showing the mesh generation result of the blast furnace body modeling area in an embodiment of the present invention;

[0180] Figure 5 This is a diagram showing the geometric parameters of a blast furnace in an embodiment of the present invention;

[0181] Figure 6 This is a circumferential temperature distribution diagram at different heights on the inner surface of the blast furnace in an embodiment of the present invention;

[0182] Figure 7 This is a longitudinal distribution curve of the blast furnace inner surface temperature at 0°C in an embodiment of the present invention;

[0183] Figure 8 This is a vector diagram of the gas velocity in the blast furnace body in an embodiment of the present invention;

[0184] Where (a) represents the original number of grid cells, and (b) represents the number of grid cells to be adjusted;

[0185] Figure 9 This is a pseudo-color image of the gas velocity in the blast furnace body in an embodiment of the present invention;

[0186] Figure 10 This is a pseudo-color image of the blast furnace body temperature distribution in an embodiment of the present invention;

[0187] Where (a) represents the gas phase temperature distribution and (b) represents the solid phase temperature distribution. Detailed Implementation

[0188] 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.

[0189] A mechanism- and data-based intelligent modeling method for blast furnaces includes the following steps:

[0190] Step 1: Establish a model of the temperature field distribution on the inner surface of the furnace wall;

[0191] Step 1.1: Obtain an overview of the furnace wall temperature;

[0192] For blast furnaces with their high-temperature and high-pressure internal environment, the furnace wall, as the physical framework, must not only withstand the high temperature and pressure at the furnace's inner edges but also ensure the stability of the internal temperature. Furthermore, to isolate the blast furnace from the external environment, the furnace wall must also provide cooling and insulation. Considering practicality and economy, the blast furnace wall is generally composed of multiple materials from the inside out. Depending on the materials, the furnace wall typically has a four-layer structure: from the outside in, the shell, the filling layer, the cooling wall, and the refractory bricks. To better observe and assess the blast furnace's condition, several thermocouples are installed at different heights along the furnace wall. Thermocouples at the same height are evenly distributed circumferentially between the cooling pipes and the inner wall of the blast furnace; some are embedded in the refractory bricks, while others are located within the cooling wall. Data from the thermocouples measures the temperature at corresponding locations on the furnace wall. Since the heat originates from inside the blast furnace, these temperatures are generally positively correlated with the internal temperature. For blast furnaces, these thermocouples are numerous, evenly distributed, and closely related to the internal temperature. This invention selects the area between the inner wall and the cooling pipes inside the cooling wall as the research area for modeling. Before establishing the model, it is necessary to determine the physical parameters of the furnace wall and the thermocouple temperature parameters inside the furnace wall based on existing data.

[0193] Step 1.2: Establish and discretize the differential equation of heat conduction of the blast furnace wall;

[0194] The thermal conductivity differential equation is derived from the solid energy conservation law and Fourier's law for a stable interior furnace wall without internal heat sources:

[0195]

[0196] Where λ is the thermal conductivity, T is the temperature, and x, y, z are the coordinate directions of the blast furnace.

[0197] Since the furnace wall has a circular cylindrical structure, it is more convenient to solve the thermal conductivity differential equation by transforming it into a cylindrical coordinate system. The thermal conductivity differential equation in the rectangular coordinate system in equation (1) is obtained by transforming x = rcosθ, y = rsinθ, z = h:

[0198]

[0199] Where r is the radius of the circular cross-section of the blast furnace wall, θ is the angle of the circular cross-section of the blast furnace wall, and h is the height of the blast furnace.

[0200] Since thermocouples are distributed layer by layer along the longitudinal direction in the blast furnace, the blast furnace wall is divided into several cross-sections, forming several two-dimensional cross-sectional heat conduction models. For each two-dimensional cross-sectional heat conduction model, ignoring the longitudinal direction, we obtain:

[0201]

[0202] Where r0 is the radius of the inner ring of the two-dimensional cross-sectional heat conduction model, and r2 is the radius of the outer ring of the two-dimensional cross-sectional heat conduction model.

[0203] Since the heat conduction model of the blast furnace wall consists of partial differential equations, which are difficult to solve analytically, the finite difference method is used to discretize the partial differential equations into algebraic equations. Because the target region is regular, it is divided into a grid with equal step sizes, and algebraic values ​​on the grid are used to replace the continuous numerical domain. The target region is divided into M steps in the radial direction and N steps in the circumferential direction. The entire target region, including the inner and outer boundaries, contains N*(M+1) discrete points. Figure 1 As shown. Black dots represent the current temperature point, gray dots represent the temperature points associated with it, and white dots represent other temperature points. Then, equation (3) is discretized using central difference, resulting in the discretized heat conduction differential equation as follows:

[0204]

[0205] Where i and j are the subscripts of the radial and circumferential discrete points, respectively, Δr is the radial step size of the two-dimensional cross-sectional heat conduction model, Δθ is the circumferential step size of the two-dimensional cross-sectional heat conduction model, and T... i,j Let r be the temperature of the grid point with index i,j after discretization. i Let i be the radius of the grid point with index i.

[0206] Step 1.3: Determine the boundary conditions of the blast furnace wall;

[0207] After describing the heat conduction process of a solid using the heat conduction differential equation, if you want to solve the heat conduction differential equation to obtain the temperature distribution, you also need to analyze the boundary conditions of the model, including the physical boundary and the corresponding temperature distribution. The most important of these are the temperature characteristics of the two sides of the modeling region and the heat transfer situation.

[0208] For the blast furnace wall, the inner surface of the wall, which serves as the inner boundary, is also the outer boundary of the reaction bed inside the furnace, and its temperature is the same as the temperature of the reaction bed inside the furnace, which is a first-type boundary condition. The junction between the cooling wall and the embedded cooling pipes, which serves as the outer boundary, is the contact between the cooling wall and the cooling water, and can be regarded as convective heat transfer between the cooling wall and the cooling water, which is a third-type boundary condition. The boundary conditions are discretized.

[0209] Including the inner surface boundary of the furnace wall:

[0210] T 1,j =T in,j , r=r0,0≤j≤N (5)

[0211] Boundary at the junction of refractory bricks and cooling wall:

[0212]

[0213] Discretized:

[0214]

[0215] Cooling wall outer surface boundary:

[0216]

[0217] Discretized:

[0218]

[0219] In the formula, T in This refers to the temperature distribution at the inner boundary, i.e., the temperature distribution on the inner surface of the blast furnace wall; T out This refers to the temperature distribution at the outer boundary, specifically the temperature distribution at the junction of the blast furnace cooling wall and the cooling pipes; T CW T represents the average temperature of the cooling water in the cooling pipes. CW =(T CW,in +T CW,out ) / 2, T CW,in T represents the temperature measured by the thermocouple at the cooling water inlet. CW,out Temperature measured by thermocouple at the cooling water outlet; h x v is the convective heat transfer coefficient between the cooling wall and the cooling water. 水 Cooling water flow rate:

[0220] h x =208 + 47.5v 水 (10)

[0221] r0, r1, and r2 are the physical parameters of the blast furnace wall. r0 is the inner diameter of the inner ring of the two-dimensional cross-sectional heat conduction model, which is the radius of the inner wall of the blast furnace. r1 is the radius at the junction of the refractory brick and the cooling wall. r2 is the radius of the outer ring of the two-dimensional cross-sectional heat conduction model, which is the radius at the location of the cooling pipe. λ1 is the thermal conductivity of the refractory brick, and λ2 is the thermal conductivity of the cooling wall.

[0222] Step 1.4: Use the optimized PSO algorithm to solve for the circumferential one-dimensional temperature distribution on the inner surface of the furnace wall;

[0223] At this point, the differential equations for heat conduction in the blast furnace wall and the boundary conditions have been determined and discretized. A method for solving the internal temperature distribution of the furnace wall based on specific boundary conditions has been presented. The following section uses the above model and the previously described thermocouple temperatures inside the furnace wall, combined with an algorithm, to calculate the one-dimensional circumferential temperature distribution on the inner surface. Since there are multiple layers of thermocouples in the blast furnace from bottom to top, this model requires multiple solutions, placing high demands on the speed of the optimization algorithm. Comparing optimization algorithms, it can be found that genetic algorithms and similar algorithms have high randomness in crossover and mutation, are slow, and are difficult to interfere with, while particle swarm optimization (PSO) algorithms are easier to interfere with due to particle motion and have memory properties in iteration. Therefore, this study chooses PSO as the basis and improves it to solve the inverse thermodynamic problem of this research.

[0224] The Particle Swarm Optimization (PSO) algorithm was first proposed by Eberhart and Kennedy.

[0225] The basic formula for the PSO algorithm is:

[0226]

[0227] In the formula, i is the i-th particle, i = 1, 2, ..., N; ω is the inertia weighting factor; V i (t) represents the velocity of the i-th particle in the t-th iteration; V i (t+1) is the velocity of the i-th particle in the (t+1)-th generation; X i (t) represents the position of the i-th particle in the t-th generation; P i (t) represents the optimal position of a single particle in generation t; P g (t) represents the global optimal position of the particle swarm at generation t, C1 and C2 are acceleration coefficients, and R1 and R2 are random numbers uniformly distributed in [0,1].

[0228] An excellent inversion recognition algorithm requires strong global search capabilities in the initial stage and excellent local search capabilities in the final stage. However, the inertial weight factor ω and acceleration coefficients C1 and C2 in the PSO method are constant, which limits the performance of the PSO method. Larger inertial weight factors ω and acceleration coefficients C1 and smaller acceleration coefficients C2 can enhance the global search capability. In contrast, smaller inertial weight factors ω and acceleration coefficients C1 and larger acceleration coefficients C2 maintain the algorithm's local search capability. Therefore, the inertial weight factor and acceleration coefficients in the PSO method are redefined as follows:

[0229]

[0230] In the formula ω ini Let ω be the initial value. fin The final value of ω; M is an integer representing the number of iterations, C 1,ini and C 1,fin These represent the initial and final values ​​of C1, respectively; C 2,ini and C 2,fin Let ω represent the initial and final values ​​of C2, respectively. ini C 1,ini and C 2,fin For a larger value, ω fin C 1,fin and C 2,ini It is a relatively small value.

[0231] Traditional PSO algorithms often get trapped in local optima, especially with a small number of particles. Our model prioritizes efficiency and has a small population size, making it prone to local optima. To overcome this, we introduce a novel evolutionary strategy to increase the randomness of particle evolution. When the particle diversity value is small and the distribution is relatively concentrated, the particles can escape the vicinity of local optima and perform a global optimum search. The particle diversity(t) is defined as:

[0232]

[0233] In the formula: L max and L min represents the maximum and minimum values ​​of the search space, respectively; j represents the parameters in each particle (32 in total); and i represents the particle number.

[0234] The search strategy for the next particle time step is selected based on the differences:

[0235] (1)diversity(t)>ξ||R3>a:

[0236]

[0237] (2)diversity(t)≤ξ&R3≤a:

[0238]

[0239] In the formula: R1, R2, R3, R4, and R5 represent random numbers uniformly distributed in [0,1]; ξ represents the constraint value of diversity(t); and a is a parameter that controls the initialization probability of particles.

[0240] Using the optimized PSO algorithm described above, the difference between the actual and calculated values ​​of the thermocouples inside the furnace wall is used as the objective function to solve the two-dimensional cross-sectional heat conduction model of the blast furnace wall, which is composed of the heat conduction differential equation obtained in step 1.2 and the boundary conditions obtained in step 1.3, to obtain the one-dimensional circumferential temperature distribution on the inner surface of the multi-layer furnace wall.

[0241] Step 1.5: Solve for the one-dimensional longitudinal temperature distribution on the inner surface.

[0242] Although blast furnace structures are physically interconnected, their structures, internal chemical reactions, material movement, and material morphology vary significantly. The rate of temperature change along the axial direction of the inner surface also differs considerably. Therefore, this paper considers the number of data point layers and the longitudinal similarity of the blast furnace, dividing it into three parts: the throat and upper body, the lower body, and the waist and belly. Figure 2 As shown. Then, the temperature distribution of each section of the blast furnace in the longitudinal direction is fitted using the cubic curve calculation formula:

[0243] T w (θ j ,h)=β 0,j +β 1,j h+β 2,j h 2 +β 3,j h 3 , j=1,2,...,N (16)

[0244] Since the number of layers in each part is inconsistent, in order to make full use of the data of each layer, Equation (17) is used for fitting to obtain the one-dimensional longitudinal temperature distribution at different angles on the inner surface of the blast furnace wall.

[0245]

[0246] In the formula: h represents the height of the blast furnace at different locations. i Let T be the height of the i-th layer. w (θ j h) represents the temperature at different locations within the blast furnace; different angles and heights correspond to different positions on the inner surface. β 0,j β 1,j β 2,j β 3,jrepresents the coefficients to be fitted to the cubic curve, and n represents the number of layers in the current part.

[0247] Thus, the one-dimensional longitudinal temperature distribution at all angles of the inner surface of the furnace wall is obtained. Combining these, the temperature field distribution of the inner surface of the furnace wall is obtained.

[0248] Step 2: Establish a two-dimensional temperature field distribution model of the blast furnace body;

[0249] Step 2.1: Use fluid theory and the Navier-Stokes equations to describe the blast furnace gas, analyze the flow, heat transfer and chemical reactions of the gas and solid phases, and establish a two-dimensional temperature field model of the blast furnace body by establishing partial differential equations describing four variables: pressure difference in the blast furnace body, gas velocity, gas temperature and solid temperature. The two-dimensional temperature field model of the blast furnace body is divided into three sub-models according to the coupling degree: a gas flow model describing gas pressure and velocity; a gas temperature model describing gas temperature; and a solid temperature model describing solid temperature.

[0250] (1) The gas flow model mainly represents gas flow and the coupling relationship between pressure and velocity, requiring the use of the continuity equation and the momentum conservation equation. The gas flow model consists of the continuity equation and the momentum conservation equation considering momentum transfer:

[0251]

[0252]

[0253]

[0254] In the formula: x and y represent the radial and axial directions, respectively; u and v represent the radial and axial velocities, respectively; p represents pressure; t represents time; and β represents the diffusion coefficient. μ is the kinetic viscosity coefficient, ρ is the gas density, and f1 and f2 are source term coefficients.

[0255]

[0256] In the formula η g ε is the gas viscosity; s The porosity of the furnace charge; ψ s d is the shape factor of the solid particles; s The diameter of the solid particle;

[0257] (2) The gas temperature model mainly considers the distribution of gas temperature, and is represented by the energy conservation equation considering energy transfer and chemical reactions:

[0258] The gas temperature model considering energy transfer is as follows:

[0259]

[0260] In the formula: T g T represents the gas temperature. s Let S be the solid temperature, S be the effect of the source term on the gas temperature model, and α be the diffusion term coefficient. k is the thermal conductivity according to Fourier's law, c v γ is the specific heat, and γ is the heat transfer coefficient.

[0261]

[0262] In the formula: A g-s k is the contact area. g d is the thermal conductivity of the gas; s Pr is the average particle size; Re is the Reynolds number; Pr is the average particle size. g The Planck number represents the convective heat transfer number of the gas. Represents the energy effect of indirect reduction reactions on gases, unit: W / m³ 3 ; Represents the effect of direct reduction reaction on gases, unit: W / m³ 3 .

[0263] (3) The solid temperature model is represented by the heat conduction equation (derived form of Fourier's law) considering gas-solid heat transfer and chemical reactions:

[0264]

[0265] In the formula: Represents the energy effect of indirect reduction reactions on solids, unit: W / m 3 ; The effect of direct reduction reaction on solids, unit: W / m 3 .

[0266] Step 2.2: Determine the modeling region and boundary conditions for the two-dimensional temperature field model of the blast furnace.

[0267] Modeling area such as Figure 3 As shown, this is the area between the lower part of the blast outlet and the cross-shaped temperature measurement point in the furnace throat.

[0268] The boundary conditions include:

[0269] Inlet: The air vent is the inlet boundary of the model;

[0270] Outlet: The furnace top is set as a pressure outlet, and the parameters of this boundary, except for pressure, are obtained by interpolation from inside the modeling region;

[0271] Furnace wall: The flow at the wall surface is set to a no-slip boundary, and the furnace wall is impenetrable; therefore, the furnace wall velocity is set to 0. Heat transfer is set to convective heat transfer, i.e., the gas exchanges heat with the furnace wall via convection.

[0272] Symmetry axis: The blast furnace is a symmetrical model. The gradient of the vertical velocity at the symmetry axis is set to 0, and the magnitude of the normal velocity is 0.

[0273] Step 3: Solve the two-dimensional temperature field model of the blast furnace body

[0274] Step 3.1: Mesh the modeling region of the two-dimensional temperature field distribution model of the blast furnace body;

[0275] Modeling area such as Figure 3 As shown, the area from the lower part of the blast furnace inlet to the crosshair temperature measurement point in the furnace throat was analyzed. Body-Fitted Coordinates (BFC) mesh generation technology was used to obtain the body coordinates for this region. Elliptic partial differential equations were used to generate the mesh for the modeling region. The convergence of the mesh iteration was determined based on the maximum difference between the physical plane coordinates of the current iteration and the previous iteration, as shown in the following equation:

[0276]

[0277] Where e is the convergence error. This represents the maximum difference between the current round and the previous round's physical plane coordinates x-axis. This represents the maximum difference between the physical plane coordinates y-axis of the current round and the previous round.

[0278] To obtain a higher quality mesh, a BFC mesh is generated by solving the Poisson equation. Let (x,y) be the physical plane coordinate system and (ξ,η) be the computational plane coordinate system. After transformation, the Poisson equation is solved using the computational plane to obtain the equations for the physical plane:

[0279]

[0280] In the formula: J, α, β, γ are coefficients:

[0281]

[0282] P(x,y) and Q(x,y) are the source terms. This study uses the calculation form established by Wei Wenli et al.:

[0283]

[0284] The formulas for calculating φ and ψ on the boundary are as follows:

[0285]

[0286] After obtaining φ and ψ on the boundary, the values ​​of φ and ψ at the internal points are obtained by linear fitting. Then, P and Q of all grid points are obtained according to equation (28), and J, α, β, and γ are calculated simultaneously to obtain the relationship between the physical plane and the computational plane. The adaptive grid obtained in this embodiment is as follows: Figure 4 As shown, the modeling area is divided into 60 steps radially and 271 steps vertically, for a total of 16,260 grids. The average step size of all grids is about 100 mm, which can not only accommodate multiple particles, but also reduce the computational delay while ensuring the accuracy of the model.

[0287] Step 3.2: Solve the two-dimensional temperature field of the blast furnace body using the SIMPLE algorithm;

[0288] Since pressure and velocity are solved based on a gas flow model, which consists of momentum and continuity equations, and these equations are coupled, direct discrete solutions are not possible. The SIMPLE algorithm is used to solve the two-dimensional temperature field of the blast furnace. The SIMPLE algorithm employs a "first assume, then correct" approach: after solving for the velocity using the momentum equation based on the assumed pressure distribution, iterative corrections are made using a pressure correction equation to ensure that the velocity satisfies both the momentum and continuity equations simultaneously. The derivation of the pressure correction equation is as follows.

[0289] First, the velocity before correction is obtained from the momentum equation, explained using radial velocity u; v is similarly derived from u. Discretizing the momentum equation, we have:

[0290]

[0291] In the formula The velocity term before calculation point correction; This is a coupling term for the velocities of adjacent points before correction; b e For source terms; This is the gradient term for the pressure correction value. Where a e Velocity before calculation point correction The coefficient, similarly, a nb and A e These are all coefficients of its suffix parameters. Since an interleaved grid method is used, the subscripts e / E, w / W, n / N, and s / S represent the left, right, lower, and upper boundaries of the main control unit grid, and p represents the current pressure point;

[0292] The corrected velocity and pressure satisfy the momentum equation, so:

[0293]

[0294] Where u e ′、u nb ′、P P ′ and P E ′ are the velocity and pressure correction values ​​for the calculation point and adjacent points, respectively.

[0295] Therefore:

[0296]

[0297] The SIMPLE algorithm makes a bold assumption that it ignores the coupling effect of the surrounding mesh velocity, and only considers the gradient term of the pressure correction value:

[0298] a e u e ′=A e (P P ′-P E ′) (33)

[0299] Therefore, the velocity correction equation is:

[0300]

[0301] In the formula:

[0302]

[0303] At the same time, the corrected pressure and velocity need to satisfy the continuity equation, therefore:

[0304] (u e -u w )Δy+(v n -v s Δx=0 (36)

[0305] In the formula: Δx is the radial step size, and Δy is the axial step size.

[0306] Substituting equation (34) into equation (36), we get the pressure correction equation:

[0307] a P P P ′=a E P E ′+a W P W ′+a N P N ′+a S P S ′+b P (37)

[0308] In the formula: subscripts E, W, N, and S represent the neighboring points in the left, right, down, and up directions of the calculation point, and coefficient a P a E a W a N and a S for:

[0309]

[0310] b PThis reflects the distance to convergence, and the internal iteration process is the process of reducing this value.

[0311]

[0312] The SIMPLE algorithm still has other details, such as the sub-relaxation of the pressure correction. Because the coupling effect of neighboring velocities is ignored during the derivation of the pressure correction, the obtained pressure correction may exceed the target value. Therefore, sub-relaxation of the pressure correction value is necessary. Generally, the relaxation factor is ω. P =0.8.

[0313] Step 3.3: Use the POD order reduction algorithm to accelerate the solution speed of the two-dimensional temperature field model of the blast furnace body;

[0314] Orthogonal intrinsic decomposition (POD) is a method for statistical analysis of vector data. It accelerates computational efficiency by reducing the dimensionality of matrices.

[0315] Step 3.3.1 Constructing the POD base

[0316] Assuming the two-dimensional temperature field model of the blast furnace body has a total time step count of T, N steps in the horizontal direction and M steps in the vertical direction, and a total number of nodes of m; firstly, the parameters of the SIMPLE algorithm solution process with L time steps are taken, including radial velocity u, axial velocity v, pressure p, and gas temperature T. g and solid temperature T s The distributions, 0 < L ≤ m, m = MN, respectively form the following five matrices A. u =(u i ) m*L A v =(v i ) m*L A p =(p i ) m*L A tg =(tg i ) m*L and A ts =(ts i ) m*L tg and ts are used to represent the gas temperature and solid temperature variables, respectively, instead of t. g and t s .

[0317]

[0318]

[0319] because and The dimension m is much larger than and The dimension L is such that, for ease of calculation and to speed up the solution, we take... and The eigenvectors and eigenvalues ​​are shown in Table 1:

[0320] Table 1. Eigenvectors and Eigenvalues ​​of Matrices

[0321] Table 1 Matrix eigenvectors and eigenvalues

[0322]

[0323]

[0324]

[0325] According to Table 1 and Equation (4.11), we obtain and eigenvectors.

[0326] The POD basis is constructed by taking several column vectors from the feature matrix formed by the feature vectors, as shown in Table 2.

[0327] Table 2 is the base table for POD.

[0328]

[0329] Step 3.3.2: Define and solve for the current POD eigenvalues;

[0330] Assume the solution to the temperature model of the entire blast furnace body is: and Then, they converge to

[0331]

[0332] Define the POD feature value as:

[0333]

[0334] Then, the POD feature value of the previous L time period can be calculated based on the current data of the previous L time period and the corresponding POD basis.

[0335] Step 3.3.3: Solve for the POD eigenvalues ​​at the next time step;

[0336] Based on the two-dimensional temperature field model of the blast furnace body, we can obtain...

[0337]

[0338] in This represents the set of differences between the parameters of the current time step and the previous time steps, which can be obtained from the discretized two-dimensional temperature field model of the blast furnace body.

[0339] According to equations (40) and (41), we obtain

[0340]

[0341] According to equations (42) and (43):

[0342]

[0343] in

[0344]

[0345] Then the POD eigenvalue at the next moment can be obtained according to equation (44).

[0346] Step 3.3.4: Obtain the parameters for the next moment, including axial velocity, longitudinal velocity, pressure, gas temperature, and solid temperature distribution until convergence;

[0347] Equation (46) is derived from equation (41). Combining the POD eigenvalues ​​and the POD basis at the next time step, the parameter distribution at the next time step is obtained.

[0348]

[0349] Then, by continuing the solution until convergence, the target value of the blast furnace internal temperature model can be obtained.

[0350] Step 3.3.5: POD Base Update

[0351] In the above steps, use to replace This is also the main source of error in the POD method, therefore error estimation is required for the POD reduction algorithm. As time progresses, the error estimation caused by the old POD basis will become increasingly larger, therefore, the POD basis also needs to be updated. The principle for updating the POD basis was determined based on the error estimation of the POD reduction algorithm.

[0352] There is a POD error of

[0353]

[0354] have to

[0355]

[0356] According to equation (40), we have

[0357]

[0358] According to the definition of differential convergence, we have:

[0359] Under the premise of ensuring the stability of the discrete equations, according to equations (48) and (49), we get

[0360]

[0361] in and These represent the radial velocities at time steps n and n-1, radial step size i, and axial step size j, respectively. and These represent the radial velocities at time steps n and n-1, radial step size i, and axial step size j, respectively; Δt is the time step; Δx is the radial step; Δy is the axial step; and e is the axial step. n+1 and e n These represent the errors at time steps n+1 and n, respectively.

[0362] Under the condition of ensuring the stability of the equations, the following error estimate exists between the accurate solution of the two-dimensional temperature distribution model of the blast furnace and the numerical solution of the POD order reduction algorithm:

[0363]

[0364] in

[0365]

[0366] If satisfied

[0367]

[0368] Then δ≤1.

[0369] If you select M u M v M p M tg and M ts satisfy Then C n (δ)=(1+δ) n-L This is the decisive factor in determining POD base updates. Therefore, the following conclusion can be drawn:

[0370] if This necessitates using a new snapshot matrix to determine the new POD base instead of the old one.

[0371] Step 4: Obtain the physical and physicochemical parameters of the blast furnace to be tested, determine the thermocouple conditions inside the blast furnace wall, and then obtain the temperature distribution on the inner surface of the furnace wall and the two-dimensional temperature distribution of the furnace body. This can effectively predict the dynamic characteristics of the blast furnace to a certain extent and provide reference value for the operation of the blast furnace smelting process.

[0372] This embodiment uses actual on-site blast furnace data for simulation case illustration.

[0373] This embodiment takes a real-world example, using the No. 1 blast furnace of a steel company as the research object, and performs data extraction, modeling, and simulation. The blast furnace is as follows... Figure 5 As shown in the figure. The values ​​of the parameters are shown in Table 3.

[0374] Table 3 Geometric parameters of blast furnace and their specific values

[0375]

[0376] Thermocouples of varying numbers are distributed circumferentially at different heights of the blast furnace wall, and the thermocouples between different layers are evenly distributed circumferentially. This study analyzed a total of 12 layers: the furnace waist, furnace belly, furnace body, and furnace throat. The physical parameters of each layer of the blast furnace wall are shown in Table 4 below.

[0377] Table 4 Physical parameters of the thermocouple layer in the blast furnace wall

[0378]

[0379] The temperature distribution of the inner surface of the blast furnace wall in all 12 layers was obtained using the heat transfer mechanism model and optimized PSO algorithm described in this invention. The temperature distribution of the inner surface of all 12 layers is shown below. Figure 6 As shown in the figure, the numbers are arranged sequentially from top to bottom, with the top line being 1 and the bottom line being 12. The figure shows that the inner surface temperature gradually increases with increasing height, which is consistent with reality.

[0380] To obtain the longitudinal temperature distribution, similar layers were grouped together to ensure the specificity of each segment as much as possible. The resulting modeling segments are shown in Table 5.

[0381] Table 5. Segmentation Table of Blast Furnace Inner Surface Fitting

[0382]

[0383] Then, a cubic curve fitting formula is used to fit each segment, and finally the results are combined. This yields the longitudinal temperature distribution of the inner surface of the furnace wall in four directions, as shown below. Figure 7 As shown, it can be observed that due to the unevenness of the furnace charge and the unevenness of the blast flow, the state of the furnace varies at different angles. However, generally speaking, the blast furnace still has a certain degree of symmetry.

[0384] Simulation of blast furnace body temperature distribution: Data from the blast furnace at 0°C is used for modeling and simulation, with L = 200 to calculate the POD basis. First, the gas velocity vector can be obtained. Figure 8 As shown in (a), the vector diagram allows observation of the velocity direction, providing a clear visual indication of the gas flow direction. The diagram clearly shows that after the gas enters through the vent, it swirls, creating a vortex zone around the vent.

[0385] Figure 8 (a) The description of the tuyeres is relatively clear, but because the number of vector points is taken from the number of grid points, the data volume is enormous, making it difficult to observe the vector distribution of the furnace body. Therefore, by reducing the number of vector points by a factor and enlarging the vector arrows, it can be seen that after the gas swirls through the tuyeres' swirling zone, it moves towards the top of the furnace under pressure, and the velocity direction is basically along the axial direction. For example... Figure 8 As shown in (b).

[0386] Besides direction, the magnitude of gas velocity also needs to be considered. A pseudo-color plot of velocity magnitude is shown below. Figure 9 As shown, the gas enters through the tuyeres and moves towards the axis under the influence of the blast furnace's kinetic energy, then moves upwards under the influence of resistance and pressure. Simultaneously, due to the material line's shape—lower in the middle and higher at the edges—the temperature in the center of the blast furnace is higher and the porosity of the burden is larger, thus the airflow velocity in the center is higher than at the edges. The tuyeres swirl zone has extremely high velocity compared to other areas; for better observation, the gas velocity in the tuyeres swirl zone is set to 5 m / s.

[0387] The temperature distribution of the gas phase can be obtained as follows: Figure 10 As shown in (a), the temperature is highest at the tuyere. This is mainly because the tuyere is the heat source of the blast furnace; both the heat carried by the blast and the heat generated by coke combustion are reflected at the tuyere. Due to the blast velocity, the gas carries heat towards the central axis and moves towards the furnace top under pressure. During this process, heat loss occurs due to heat exchange with solids and the furnace wall, which explains the step-like temperature decrease from the tuyere towards the furnace top.

[0388] Temperature distribution of solid phase as Figure 10 As shown in (b), the temperature distribution is basically consistent with that of the gas phase. This is mainly because after the charge is fed, it rapidly heats up to the surface temperature and is heated by the rising gas during its descent, gradually heating up to an equilibrium state close to the gas phase temperature. At the same time, due to the presence of convective heat transfer, there will also be a certain temperature difference between the gas and the solid.

[0389] 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 the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A mechanism- and data-based intelligent modeling method for blast furnaces, characterized in that, Includes the following steps: Step 1: Establish a model of the temperature field distribution on the inner surface of the furnace wall; Step 1.1: Obtain an overview of the furnace wall temperature; Several thermocouples are installed at different heights on the blast furnace wall. The thermocouples at the same height are evenly distributed circumferentially between the cooling pipe and the inner wall of the blast furnace. The temperature at the corresponding position on the furnace wall is measured through the data feedback of the thermocouples, and the physical parameters of the furnace wall and the temperature parameters of the thermocouples inside the furnace wall are determined. Step 1.2: Establish and discretize the differential equation of heat conduction of the blast furnace wall; Step 1.3: Determine the boundary conditions of the blast furnace wall; Step 1.4: Use the optimized PSO algorithm to solve for the circumferential one-dimensional temperature distribution on the inner surface of the furnace wall; Step 1.5: Solve for the one-dimensional longitudinal temperature distribution on the inner surface; The blast furnace is divided into three parts: the throat and upper part of the furnace body, the lower part of the furnace body, and the waist and belly. Then, the temperature distribution of each part of the blast furnace in the longitudinal direction is fitted using a cubic curve calculation formula. (16); Since the number of layers in each part is not consistent, in order to make full use of the data of each layer, Equation (17) is used for fitting to obtain the one-dimensional longitudinal temperature distribution at different angles on the inner surface of the blast furnace wall. (17); In the formula: The height of different positions of the blast furnace. For the first The height of the layer position, Different temperatures, angles, and heights correspond to different positions on the inner surface of the blast furnace. , , , The coefficients to be fitted to the cubic curve are . This represents the number of layers in the current section. Thus, the one-dimensional longitudinal temperature distribution at all angles of the inner surface of the furnace wall was obtained. Combined together, the temperature field distribution of the inner surface of the furnace wall was obtained. Step 2: Establish a two-dimensional temperature field distribution model of the blast furnace body; Step 3: Solve the two-dimensional temperature field model of the blast furnace body; Step 3.1: Mesh the modeling region of the two-dimensional temperature field distribution model of the blast furnace body; The modeling region is located between the lower part of the tuyeres and the cross-shaped temperature measurement point in the furnace throat. An elliptic partial differential equation is used to generate the mesh for this region. Mesh convergence is determined by the maximum difference between the physical plane coordinates of the current iteration and the previous iteration, as shown in the following formula: (25); in To minimize convergence error, The physical plane coordinates of the current round and the previous round The maximum difference between the axes, The physical plane coordinates of the current round and the previous round The maximum difference between the axes; BFC meshes are generated by solving the Poisson equation. For the physical plane coordinate system, To calculate the planar coordinate system, the Poisson equation is transformed, and the equations of the physical plane are solved using the computational plane as follows: (26); In the formula: , , , For coefficients: (27); , For source terms: (28); The boundary , The calculation formula is: (29); Find the boundary , Then, linear fitting was used to obtain the interior points. , The value is then calculated using equation (28) to obtain the values ​​of all grid points. , Simultaneously calculate , , , This yields the relationship between the physical plane and the computational plane; Step 3.2: Solve the two-dimensional temperature field of the blast furnace body using the SIMPLE algorithm; The SIMPLE algorithm calculates the velocity based on the assumed pressure distribution using the momentum equation, and then uses the pressure correction equation for iterative correction so that the velocity satisfies both the momentum equation and the continuity equation. Step 3.3: Use the POD order reduction algorithm to accelerate the solution speed of the two-dimensional temperature field model of the blast furnace body; Step 4: Determine the thermocouple conditions inside the blast furnace wall based on the physical and physicochemical parameters of the blast furnace to be tested, and obtain the temperature distribution on the inner surface of the furnace wall and the two-dimensional temperature distribution of the furnace body.

2. The intelligent blast furnace modeling method based on mechanism and data according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2.1: The blast furnace gas is described using fluid theory and the Navier-Stokes equations. The flow, heat transfer, and chemical reactions of the gas and solid phases are analyzed. A two-dimensional temperature field model of the blast furnace body is established by setting up partial differential equations describing four variables: pressure difference, gas velocity, gas temperature, and solid temperature. The two-dimensional temperature field model of the blast furnace body is divided into three sub-models according to the coupling degree: a gas flow model describing gas pressure and velocity; a gas temperature model describing gas temperature; and a solid temperature model describing solid temperature. Step 2.2: Determine the modeling region and boundary conditions for the two-dimensional temperature field model of the blast furnace, where the boundary conditions include: Inlet: The air vent is the inlet boundary of the model; Outlet: The furnace top is set as a pressure outlet, and the parameters of this boundary, except for pressure, are obtained by interpolation from inside the modeling region; Furnace wall: The flow on the wall is set to a no-slip boundary, and the furnace wall is impenetrable, so the furnace wall velocity is set to 0; the heat transfer is set to convective heat transfer, that is, the gas exchanges heat with the furnace wall through convection. Symmetry axis: The blast furnace is a symmetrical model. The gradient of the vertical velocity at the symmetry axis is set to 0, and the magnitude of the normal velocity is 0.

3. The intelligent blast furnace modeling method based on mechanism and data according to claim 2, characterized in that, In step 1.2, the differential equation for heat conduction is obtained from the conservation of energy in solids and Fourier's law, which is: (1); in Thermal conductivity, For temperature, , , The coordinate direction of the blast furnace; The heat conduction differential equation in the rectangular coordinate system in equation (1) is transformed by... , , The transformation yields: (2); in The radius of the circular cross-section of the blast furnace wall is given by [reference to radius]. The angle of the circumferential cross-section of the blast furnace wall. The height of the blast furnace; Divide the blast furnace wall into several cross-sections to form several two-dimensional cross-sectional heat conduction models. For each two-dimensional cross-sectional heat conduction model, ignoring the longitudinal direction, we obtain: (3); in Let be the radius of the inner ring in the two-dimensional cross-sectional heat conduction model. Let be the radius of the outer ring of the two-dimensional cross-sectional heat conduction model; The partial differential equations are discretized into algebraic equations using the finite difference method. Since the target region is regular, it is divided into a grid with equal step sizes, and the algebraic values ​​on the grid are used to replace the continuous numerical domain. The target domain is further divided radially. Step, divided into circumferential directions Step, the entire target domain, including both inner and outer boundaries, coexists. The number of discrete points is then determined, and equation (3) is discretized by central difference to obtain the discretized heat conduction differential equation as follows: (4); in , These are the subscripts of discrete points in the radial and circumferential directions, respectively. The radial step size is for the two-dimensional cross-sectional heat conduction model. The circumferential step size is the two-dimensional cross-sectional heat conduction model. The subscript after discretization is The temperature of the grid points, For subscript The radius of the grid points.

4. The intelligent blast furnace modeling method based on mechanism and data according to claim 3, characterized in that, The boundary conditions mentioned in step 1.3 include: Inner surface boundary of furnace wall: (5); Boundary at the junction of refractory bricks and cooling wall: (6); Discretized: (7); Cooling wall outer surface boundary: (8); Discretized: (9); In the formula, This refers to the temperature distribution at the inner boundary, i.e., the temperature distribution on the inner surface of the blast furnace wall. This refers to the temperature distribution at the outer boundary, i.e., the temperature distribution at the junction of the blast furnace cooling wall and the cooling pipes. The average temperature of the cooling water in the cooling pipes. , The temperature reading measured by the thermocouple at the cooling water inlet. The temperature measured by the thermocouple at the cooling water outlet; The convective heat transfer coefficient between the cooling wall and the cooling water is denoted as . Cooling water flow rate: (10); , , These are the physical parameters of the blast furnace wall, among which... This refers to the inner diameter of the inner ring in the two-dimensional cross-sectional heat conduction model, which is also the radius of the blast furnace inner wall. The radius at the junction of the refractory brick and the cooling wall. This is the radius of the outer ring of the two-dimensional cross-sectional heat conduction model, which is also the radius at the location of the cooling pipe; The thermal conductivity of the refractory brick. is the thermal conductivity of the cooling wall.

5. The intelligent blast furnace modeling method based on mechanism and data according to claim 4, characterized in that, The basic formula of the PSO algorithm described in step 1.4 is: (11); In the formula, For the first Particle number 1 ; Inertia weighting factor; For the first The iteration of particle number 1 The speed of generation; For the first The number of particles The speed of generation; For the first The number of particles The position of the generation; For the first The optimal position of a single particle in time; For the first The global optimal position of the time-series particle swarm. , The acceleration coefficient; , for A uniformly distributed random number; The inertia weighting factor and acceleration coefficient in the PSO method have been redefined as follows: (12); In the formula for initial value, for The final value; The integer represents the number of iterations. and They represent The initial and final values; and They represent Let the initial and final values ​​be . , and For larger values, , and The smaller the value; Particle differences Defined as: (13); In the formula: and These represent the maximum and minimum values ​​of the search space, respectively. This represents the parameters in each particle. Indicates the particle number; The search strategy for the next particle time step is selected based on the differences: (1) : (14); (2) : (15); In the formula: , , , , express A uniformly distributed random number; express Limits; Parameters that control the initialization probability of particles; Using the optimized PSO algorithm described above, the difference between the actual and calculated values ​​of the thermocouples inside the furnace wall is used as the objective function to solve the two-dimensional cross-sectional heat conduction model of the blast furnace wall, which is composed of the heat conduction differential equation obtained in step 1.2 and the boundary conditions obtained in step 1.3, to obtain the one-dimensional circumferential temperature distribution on the inner surface of the multi-layer furnace wall.

6. The intelligent blast furnace modeling method based on mechanism and data according to claim 5, characterized in that, In step 2.1: (1) The gas flow model consists of a continuity equation and a momentum conservation equation that takes momentum transfer into account: (18); (19); (20); In the formula: , These are the radial and axial directions, respectively. , These are radial and axial velocities, respectively. For pressure, For time, Here are the diffusion term coefficients: , The kinetic viscosity coefficient, For gas density, , For source term coefficients: (21); In the formula For gas viscosity; Porosity of the furnace charge; The shape factor of the solid particles; The diameter of the solid particle; (2) The gas temperature model is represented by the energy conservation equation, taking into account energy transfer and chemical reactions: The gas temperature model considering energy transfer is as follows: (22); In the formula: For gas temperature, For solid temperature, The influence of the source term on the gas temperature model, Here is the diffusion term coefficient. , The thermal conductivity is given by Fourier's law. For specific heat, Heat transfer coefficient: (23); In the formula: Contact area; is the thermal conductivity of the gas; Average particle size; It is the Reynolds number; For the convective heat transfer of the gas, Prandtl number. Represents the energy effect of indirect reduction reactions on gases, in units of: ; Represents the effect of direct reduction reaction on gases, unit: ; (3) The solid temperature model is expressed by considering gas-solid heat transfer and chemical reactions using the heat conduction equation: (24); In the formula: Represents the energy effect of indirect reduction reactions on solids, in units of: ; Represents the effect of direct reduction reaction on solids, unit: .

7. The intelligent blast furnace modeling method based on mechanism and data according to claim 6, characterized in that, Step 3.2 specifically involves: First, the velocity before correction is obtained from the momentum equation, expressed as radial velocity. To explain, and Similarly, discretizing the momentum equation yields: (30); In the formula The velocity term before calculation point correction; This is a coupling term for the velocities of adjacent points before correction. For source terms; For the pressure correction gradient term; where Velocity before calculation point correction The coefficient, similarly, and All of these are coefficients of their suffix parameters. The subscripts e / E, w / W, n / N, and s / S represent the left, right, lower, and upper boundaries of the main control unit grid, and p represents the current pressure point. The corrected velocity and pressure satisfy the momentum equation, so: (31); in , , and These are the velocity and pressure correction values ​​for the calculation point and adjacent points, respectively. Therefore: (32); Considering only the gradient term of the pressure correction value, we have: (33); Therefore, the velocity correction equation is: (34); In the formula: (35); At the same time, the corrected pressure and velocity need to satisfy the continuity equation, therefore: (36); In the formula: Radial step size, The axial step size; Substituting equation (34) into equation (36), we get the pressure correction equation: (37); In the formula: subscripts E, W, N, and S represent the neighboring points in the left, right, down, and up directions of the calculation point, and coefficients , , , and for: (38); It reflects the distance to convergence; (39)。 8. The intelligent blast furnace modeling method based on mechanism and data according to claim 7, characterized in that, Step 3.3 specifically includes the following steps: Step 3.3.1 Construct the POD base; Assuming the total number of time steps in the two-dimensional temperature field model of the blast furnace body is... The space is divided into horizontal and vertical directions respectively. , Step, the total number of nodes is First, the parameters for the SIMPLE algorithm solution process with L time steps are obtained, including the radial velocity. axial velocity ,pressure Gas temperature and solid temperature distributed, These form the following five matrices respectively. , , , and ,use and Represents gas temperature and solid temperature variables; , , ; , ; Seeking , , , and eigenvectors and eigenvalues; The POD basis is constructed by taking several column vectors from the eigenmatrix formed by the eigenvectors. Step 3.3.2: Define and solve for the current POD eigenvalues; Assume the solution to the temperature model of the entire blast furnace body is: , , , and ; Therefore, they converge to: (40); Define the POD feature value as: (41); According to the current front Before the data for the time period and the corresponding POD basis are obtained. POD characteristic values ​​for a given time period; Step 3.3.3: Solve for the POD eigenvalue at the next time step; based on the two-dimensional temperature field model of the blast furnace body, we get: (42); in This represents the set of differences between the parameters of the current time step and the previous time steps, obtained from the discretized two-dimensional temperature field model of the blast furnace body. According to equations (40) and (41), we get: (43); According to equations (42) and (43), we have: (44); in: (45); The POD eigenvalue at the next moment is obtained according to equation (44); Step 3.3.4: Obtain the parameters for the next moment, including axial velocity, longitudinal velocity, pressure, gas temperature, and solid temperature distribution until convergence; Equation (46) is derived from equation (41). Combining the POD eigenvalues ​​and the POD basis at the next time step, the parameter distribution at the next time step is obtained. : (46); Then continue solving until convergence to obtain the target value of the blast furnace internal temperature model. , , , and ; Step 3.3.5: POD base update; The POD error is: (47); have to: (48); According to equation (40), we have: (49); According to the definition of differential convergence, we have: Under the premise of ensuring the stability of the discrete equations, according to equations (48) and (49), we get (50); in and These represent time steps as follows: and Radial step size is axial step size is radial velocity at the location; and These represent time steps as follows: and Radial step size is axial step size is radial velocity at the location; For time step; Radial step size; The axial step size, and The time steps are respectively and The error; Under the condition of ensuring the stability of the equations, the following error estimate exists between the accurate solution of the two-dimensional temperature distribution model of the blast furnace and the numerical solution of the POD order reduction algorithm: (51); in: (52); If the following conditions are met: (53); but ; If selected , , , and satisfy ,So This is the decisive factor in determining POD base updates; therefore, the following conclusions can be drawn: if Therefore, a new snapshot matrix is ​​needed to determine the new POD base to replace the old POD base.