Numerical simulation method for multi-physical phenomena of rotary kiln based on artificial intelligence reasoning

Abstract

The application discloses a rotary kiln multi-physical phenomenon numerical simulation method based on artificial intelligence reasoning, relates to the technical field of rotary kiln multi-physical phenomenon simulation, and is based on the Euler multiphase flow equation, adds a heat transfer model, a chemical dynamics reaction model, a particle growth model, a chemical reaction model and the like, establishes a three-dimensional numerical model capable of accurately simulating the multi-physical and chemical phenomena such as the gas-solid two-phase flow movement, combustion, NOx generation, clinker calcination and particle growth in the rotary kiln, introduces a long short-term memory neural network to build an AI model, uses the AI model to predict the latter half of the working condition, accelerates the simulation speed, and realizes long-time simulation of the rotary kiln.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of rotary kiln multi-physical phenomenon simulation, and particularly relates to a rotary kiln multi-physical phenomenon numerical simulation method based on artificial intelligence reasoning. BACKGROUND

[0002] The rotary kiln is the core equipment of the new dry cement sintering system, and the flame temperature, flame length, air supply quantity and material liquid phase quantity in the rotary kiln have a great influence on the movement of the material. Since the temperature in the rotary kiln is extremely high, the maximum temperature can reach about 1800 DEG C, and the gas contains a large amount of dust, and there is no good experimental monitoring method to monitor the internal material movement, chemical reaction and particle enlargement during the calcination of the clinker in the rotary kiln during operation; that is, the internal rotary kiln is still a "black box" for research and design personnel in industry.

[0003] In order to understand the temperature distribution, material movement and chemical reaction in the rotary kiln, numerical simulation is the most widely used and most economical method for researchers.

[0004] The patent with the publication number CN117875097A discloses a rotary kiln aluminum ash treatment simulation simulation method established by COMSOL, which establishes a two-dimensional axisymmetric geometric model. Secondly, the geometric model area is divided according to the component body of the rotary kiln. Then, the material parameters, boundary values and initial parameters are set, and the grid is divided. Finally, the calculation solution is obtained, and the rotary kiln aluminum ash treatment simulation result is obtained.

[0005] The patent with the publication number CN112380738A discloses a cement rotary kiln combustion field reconstruction error compensation and optimization method, storage medium and system. The method first establishes a numerical simulation model of the cement rotary kiln, extracts numerical simulation data and uses a neural network to obtain the relationship between the combustion efficiency and the boundary conditions, extracts the characteristic parameters of the combustion efficiency, uses the extracted characteristic parameters as error compensation modeling, and uses an online optimization algorithm to perform error compensation reconstruction on the cement rotary kiln site, and realizes online real-time optimization.

[0006] The patent with the publication number CN106570244B discloses a one-dimensional mathematical model established in the cement rotary kiln, which includes a heat transfer model, a material bed movement model and a clinker mineral chemical reaction model. The field production data are collected for the one-dimensional model boundary, and the changes of the flue gas temperature, material temperature, material bed height, wall surface temperature, material chemical composition and clinker mineral content in the rotary kiln with the length of the rotary kiln are calculated.

[0007] At present, the existing technology mainly focuses on the simulation of a single phenomenon, and cannot effectively simulate the interaction of 3D multi-physical fields and chemical phenomena in the actual operation of the rotary kiln. Therefore, a solution is provided. SUMMARY

[0008] The present application aims to solve at least one of the technical problems existing in the prior art; for this purpose, the present application proposes a rotary kiln multi-physical phenomenon numerical simulation method based on artificial intelligence reasoning.

[0009] The rotary kiln multi-physical phenomenon numerical simulation method based on artificial intelligence reasoning, the method specifically comprises the following steps:

[0010] Step one: use CAD software to draw the 3D geometric file of the rotary kiln, divide the grid, collect the kiln head temperature, kiln head pressure, kiln tail temperature, kiln tail pressure, coal quantity, primary air speed and temperature, secondary air speed and temperature, and the content data of each component in the kiln tail exhaust, and define the boundary conditions;

[0011] Step two: build a numerical simulation model, the numerical simulation model is based on the Euler two-phase flow basic model, and the remaining models will be built on the Euler two-phase flow model, the Euler two-phase flow basic model includes three basic equations of continuity equation, momentum equation and energy equation;

[0012] Add a sub-model in the numerical simulation model, the sub-model includes:

[0013] Coal combustion simulation model, used to simulate the whole process of coal combustion;

[0014] Multiphase flow transmission and heat transfer simulation model, used to simulate the movement law and heat exchange law of the burning material particles in the rotary kiln;

[0015] Clinker calcination simulation model, used to simulate the evolution of mineral components in the burning material under the action of material transmission, heat transfer and mass transfer;

[0016] Particle growth simulation model, used to simulate the situation that the particle size increases in the calcination process of the burning material, affecting the movement law and heat exchange law of the burning material particles;

[0017] Pollutant generation simulation model, used to analyze the generated pollutants in the whole process;

[0018] Complete the building of the 3D numerical simulation model of the rotary kiln.

[0019] Compared with the prior art, the present application has the following advantages:

[0020] Based on the Euler multiphase flow equation, the present application adds heat transfer model, chemical kinetics reaction model, particle growth model, chemical reaction model, etc., to establish a three-dimensional numerical model that can accurately simulate the movement, combustion, NOx generation, clinker calcination, particle growth and other multi-physical and chemical phenomena of the gas-solid two-phase flow in the rotary kiln; and introduce a long-short term memory neural network to build an AI model, use the AI model to predict the latter half of the working condition, speed up the simulation, and realize long-time simulation of the rotary kiln. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The technical roadmap of the rotary kiln multi-phenomenon simulation of the present application is shown in the figure;

[0022] Figure 2 The eigenvalue orthogonal decomposition schematic diagram of the sample data of the present application is shown in the figure;

[0023] Figure 3 The detailed structure diagram of the LSTM of the present application is shown in the figure;

[0024] Figure 4 The overall neural network diagram of the AI model of the present application is shown in the figure. DETAILED DESCRIPTION

[0025] The technical solutions of the present application will be described clearly and completely in combination with the embodiments below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0026] Please refer to Figures 1-4 The present application provides a rotary kiln multi-physical phenomenon numerical simulation method based on artificial intelligence reasoning, which specifically comprises the following steps:

[0027] S1: preparation work;

[0028] Draw the 3D geometric file of the rotary kiln by using the CAD software, perform mesh division, collect the kiln head temperature, kiln head pressure, kiln tail temperature, kiln tail pressure, coal feeding amount, primary air speed and temperature, secondary air speed and temperature, and the content data of each component in the kiln tail exhaust gas, and define the boundary conditions;

[0029] S2: numerical model building;

[0030] (1) basic equation: the numerical simulation model of the present patent is based on the Euler two-phase flow basic model, and the remaining models will be built on the Euler two-phase flow model. The Euler two-phase flow basic model includes three basic equations of continuity equation, momentum equation and energy equation;

[0031] The Euler double-fluid basic model in the present patent has three phases:

[0032] The first phase is the gas phase, and the first phase is also the main phase, which contains various gas components such as:

[0033] O2, CO2, N2, NO, CO, H2O(g), HCN, CH4, C2H4, C2H6, and H2, a total of 11 kinds of gas components;

[0034] The second phase is a coal powder solid particle, and includes four components: fixed carbon, volatile matter, ash, and H2O(l);

[0035] The third phase is a calcined material particle, and includes six components: CaCO3, CAO, C2S, C3S, C3A, and C4AF;

[0036] The Euler-Euler two-phase flow continuity equation is:

[0037]

[0038] In the formula, α q is a volume fraction of the q phase, is a q phase velocity, is a mass flow transferred from the p phase to the q phase, is a mass flow transferred from the q phase to the p phase, S q is a source phase.

[0039] The momentum control equation is:

[0040]

[0041] In the formula, k

[0042]

[0043] In the formula, μ q and λ q are a shear viscosity and a volume viscosity of the q phase, F q is an external force, F lift,q is a lift force, is a turbulent dissipation force, is a virtual mass force, Rpq is an interaction force between phases, is a phase velocity;

[0044] The energy control equation is:

[0045]

[0046] In the formula, k eff,q is a thermal conductivity, S q is a source term, Q pq is a heat exchange amount of the p phase and the q phase, h q is an enthalpy of the q phase, p op is a reference atmospheric pressure, and p is a gauge pressure;

[0047] Further, on the basis of the Euler two-fluid model, the simulation content of the clinker calcination process in the rotary kiln needs to add the following five sub-models:

[0048] (1) a coal powder combustion simulation model;

[0049] (2) Simulation model of multiphase flow transmission and heat transfer;

[0050] (3) Simulation model of clinker calcination;

[0051] (4) Particle growth simulation model;

[0052] (5) Pollutant generation simulation model, the main pollutant being NOx;

[0053] The first step in the study of rotary kilns is the simulation of pulverized coal combustion, which determines the overall temperature field of the rotary kiln.

[0054] Furthermore, multiphase flow transmission and heat transfer simulation studies were conducted to investigate the movement and heat exchange patterns of the burning material particles inside the rotary kiln.

[0055] Furthermore, the calcination simulation model of the calcined material is used to study the evolution of the internal mineral composition of the calcined material under the material transmission / heat transfer / mass transfer behavior;

[0056] Furthermore, during the calcination process, the particles gradually grow larger, which in turn affects the particle movement and heat exchange patterns of the calcined material.

[0057] Finally, considering the generation of pollutants throughout the process, which is related to the gas temperature field and gas composition, a pollutant generation simulation model is used. The above five parts are interconnected and mutually influential. Simulating a single problem is inaccurate. Only by considering all five parts in the same model can the simulation results be realistic and reliable, and have value in guiding construction.

[0058] (2) The specific simulation model for pulverized coal combustion is as follows:

[0059] The pulverized coal combustion model involves both gaseous combustion and solid combustion models. The gaseous combustion model assumes that pulverized coal volatiles will release combustible gases, and this part uses a rapid chemical reaction model.

[0060]

[0061] Among them, R i,r It is the reaction rate (kmol / s), v″ i,r With v′ i,r ε is the stoichiometric coefficient, A and B are empirical constants, 4.0 and 0.5 respectively, YR is the reactant mass fraction, Mw,r is the reactant mole fraction, YP is the reactant mass fraction, Mw,p is the reactant mole fraction, ε is the turbulent dissipation rate, and k is the turbulence intensity.

[0062] The solid combustion model uses a reduced-core model:

[0063]

[0064] where P o2 is the partial pressure of oxygen around the particle, k fm is the film layer resistance, k am is the ash layer resistance, k rm is the surface reaction resistance coefficient of the inner core;

[0065] The gas combustion is added to the primary phase and the solid combustion is added to the secondary phase in the coal combustion model, wherein the coal combustion involves the transfer of the fixed carbon component C in the secondary phase to the component CO2 in the primary phase, which adds the heterogeneous chemical reaction module;

[0066] The multiphase flow transmission and heat transfer simulation model is specifically:

[0067] The interphase (gas-solid) transmission model uses the Huilin & Gidaspow drag force model, and the specific theory is:

[0068] The force on the particle in the air is:

[0069]

[0070] where C d is the drag coefficient, p f is the fluid density (kg / m3), u is the fluid velocity (m / s), v p is the particle velocity (m / s); there are many ways to calculate the drag coefficient, and different calculation methods represent different drag force models;

[0071]

[0072] where is the Ergun drag coefficient, is the Wen-Yu drag coefficient, and p is the weight coefficient, which is (0, 1);

[0073] The heat transfer model uses the GUNN model, which is specifically:

[0074]

[0075] The clinker calcination model is:

[0076] The clinker calcination model is added to the third phase, and there are five chemical models. The reaction type and chemical kinetic parameters of each chemical reaction model have been obtained through experiments, and the specific parameters are shown in Table 1:

[0077] Table 1 Reaction type and chemical kinetic parameters

[0078]

[0079] The particle growth model is specifically as follows:

[0080] The particle growth model uses a PBM model, which can compare and quickly realize simulation of particle growth. The PBM ignores particle breakage and agglomeration, and only considers particle growth. Meanwhile, the PBM model is solved by using a discrete method, and the PBM model is coupled with a CFD model to realize simulation of the clinker calcination process. The particle growth model is added to the third phase.

[0081]

[0082] The Hillert model is used to calculate and predict the clinker calcination model, and specifically as follows:

[0083]

[0084] where d0 is an initial particle size, A is a pre-exponential factor of particle growth, E p is a particle growth activation energy, R is a gas constant, T is a particle temperature, and time is a calcination time length after the particle enters the kiln.

[0085] The pollutant generation simulation model is specifically as follows:

[0086] The pollutant model uses a finite chemical reaction rate model, and the chemical reaction rate is as follows:

[0087]

[0088] where C j,r is a molar concentration (kmol / s), η ' i,r and η ” i,r are equivalent numbers of reactants and products, Γ represents an influence of a reversible reaction, v ” i,r and v ' i,r are stoichiometric numbers.

[0089] The chemical reaction rate constant is often an Arrhenius formula as follows:

[0090]

[0091] where A r is a pre-exponential factor, T is a temperature, β is a temperature coefficient, E r is an activation energy, and R is a gas constant.

[0092] In addition, there is a catalytic reduction reaction. The reaction model can be added to a catalyst mass phase on the basis of the finite chemical rate model, that is:

[0093]

[0094] The pollutant model components are all gases, so they are added to the first phase;

[0095] So far, the 3D numerical simulation of the rotary kiln is completed.

[0096] S3: numerical simulation calculation correction;

[0097] The numerical simulation model built in S2 is calculated, and the on-site data collected in S1 is used to verify the accuracy of the built model. If there is a difference, the parameters are adjusted;

[0098] S4: long short-term memory neural network (LSTC) for long-time calculation;

[0099] The model built in S2 basically realizes the simulation of all behaviors in the clinker calcination process in the rotary kiln, but the chemical reaction size difference of the numerical model built in S2 is large, the fastest chemical reaction rate (combustion) is 1e9 times the slowest reaction rate (particle enlargement), and the stiffness of the chemical reaction is too large;

[0100] Although the use of implicit algorithms can reduce the chemical stiffness to some extent, it takes at least thirty minutes for the material to enter the kiln to the kiln, and the simulation calculation using numerical simulation for more than thirty minutes is huge. If the rotary kiln is recalculated every time the rotary kiln is adjusted, the calculation efficiency is too slow; this is the biggest difficulty in simulating the rotary kiln.

[0101] The total calculation time of the rotary kiln is 1800s, and the numerical model built in S2 is used to calculate the first 500s, and the full-field velocity field, temperature field, and three-phase gas-solid component field data of the rotary kiln are derived. Using the flow field feature analysis method based on long short-term memory neural network (LSTC), the unsupervised training method is used to fully exploit the implicit features in the time series signal, and the low-dimensional representation and feature analysis of the complex time series features in the rotary kiln flow field are performed. Get the AI model of the rotary kiln operation, and use the AI model to calculate the subsequent 1300s, which can greatly save the calculation time.

[0102] (1) Eigenvalue orthogonal decomposition of sample data

[0103] Eigenvalue orthogonal decomposition, also known as principal component analysis, is a method of statistical analysis of vector data, which can reduce the order of high-dimensional flow field data and map it to a low-dimensional orthogonal basis modal space, thereby analyzing the main features of the flow field and the corresponding basis modal coefficients. In essence, it is to make the sample variance in each dimension maximized after reducing the flow field to low dimension.

[0104] First, the sample (flow field data) calculated by simulation needs to be standardized. Let the original data be x i , i = 1, 2, 3, … r, the sample data x i is an n-dimensional vector, and r is the number of samples.

[0105]

[0106] Further, the covariance matrix of the normalized data can be obtained:

[0107]

[0108] By solving the eigenvalues of the nXn order covariance matrix, the first m order eigenvalues can be denoted as λ 1 , λ 2 , …, λ m , and the corresponding basis modal eigenvectors can be denoted as ξ 1 , ξ 2 , …, ξ m . The value of m is determined according to the proportion of the variance value of different basis modes in the total variance value, so as to ensure that the characteristic components contained in the basis modes account for more than 95% of the entire sample space.

[0109] Then the original sample data can be approximately represented as:

[0110] X = U λ ξ

[0111] In this way, the basis modal coefficient matrix and a small number of POD basis modes can be used to represent most of the information of the original sample.

[0112] (2) Convolution layer

[0113] In order to effectively extract the time sequence characteristics of different positions in the flow field, one-dimensional convolution is combined for feature extraction of time history signal. The size of all convolution kernels in the model is a large convolution kernel of 9, the number of convolution kernels is thirty-two, and the activation function uses RELU function to obtain a larger range of local time history information, which is conducive to the convergence of the model.

[0114] (3) Long short-term memory neural network LSTM;

[0115] The long short-term memory neural network can learn long-term dependent information, is suitable for processing and predicting important events with relatively long interval and delay in time sequence, and avoids the problems of gradient disappearance or gradient explosion when performing back propagation; the LSTM network structure is composed of a forgetting gate, an input gate and an output gate.

[0116] Forgetting gate: the forgetting gate takes the output ht-1 of the previous layer and the sequence data xt to be input in the current layer as input, and obtains the output ft through an activation function sigmorid. The output value of ft is in the interval [0, 1], which represents the probability of forgetting the state of the previous layer cell, 1 is completely retained, and 0 is completely forgotten.

[0117] f t = σ (Wf ·[h t-1 ,x t ]+b f )

[0118] Input gate: The input gate contains two parts, the first part uses a sigmoid activation function, and the output is it, the second part uses a tanh activation function, and the output is is the output of this layer, it is in the interval [0, 1], indicates the degree to which the information in the input is retained, indicates the new information retained by this layer.

[0119] i t = σ(W f ·[h t-1 ,x t ]+b i )

[0120]

[0121] So far, ft is the output of the forget gate, which controls the degree to which the cell state of the previous layer Ct-1 is forgotten, is the multiplication operation of the two outputs of the input gate, indicating how much new information is retained, based on which we can update the cell state of this layer Ct with new information.

[0122]

[0123] Output gate: The output gate is used to control how much of the cell state of this layer is filtered. First, a sigmoid activation function is used to obtain an ot in the interval [0, 1], then the cell state Ct is processed through a tanh activation function and multiplied by ot, which is the output of this layer ht.

[0124] o t = σ(W o [h t-1 ,x t ]+b o )

[0125] h t = o t · tanh(C t )

[0126] The main steps are as follows:

[0127] First step: forward calculation of the output value of each neuron. For LSTM, according to the algorithm introduced earlier, the calculation is performed respectively.

[0128] Second step: determine the optimization objective function. In the early training, the output value and the expected value will be inconsistent, so the error term value of each neuron is calculated to construct the loss function.

[0129] Third step: update the network weight parameters according to the gradient of the loss function. The back propagation of the LSTM error term includes two levels: one is the spatial level, which propagates the error term to the previous layer of the network. The other is the time level, which propagates the error along the time, that is, starting from the current time t, the error of each time is calculated.

[0130] Then jump to the first step, repeat the first, second and third steps until the network error is less than a given value.

[0131] The above examples are only used to illustrate the technical method of the present application and are not limited. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A numerical simulation method for multiple physical phenomena in rotary kilns based on artificial intelligence reasoning, characterized in that, The method specifically includes the following steps: Step 1: Use CAD software to draw the 3D geometric file of the rotary kiln, perform mesh generation, collect data on the kiln head temperature, kiln head pressure, kiln tail temperature, kiln tail pressure, coal feed rate, primary air velocity and temperature, secondary air velocity and temperature, and the content of each component in the kiln tail exhaust gas, and define boundary conditions. Step 2: Build a numerical simulation model. The numerical simulation model is based on the Euler two-phase flow basic model. Other models will be built on the Euler two-phase flow model. The Euler two-phase flow basic model includes three basic equations: continuity equation, momentum equation, and energy equation. Add sub-models to the numerical simulation model. These sub-models include: A pulverized coal combustion simulation model is used to simulate the entire process of pulverized coal combustion. A multiphase flow transmission and heat transfer simulation model is used to simulate the movement and heat exchange patterns of the burning material particles inside a rotary kiln. The clinker calcination simulation model is used to simulate the evolution of the mineral composition inside the calcined material under the influence of material transport, heat transfer, and mass transfer. The particle growth simulation model is used to simulate the situation where particle growth affects the particle movement and heat exchange patterns of calcined materials during the calcination process. A pollutant generation simulation model is used to analyze the pollutants generated throughout the entire process; Complete the construction of the 3D numerical simulation model of the rotary kiln.

2. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 1, characterized in that, After completing step two, the following steps are also required: Calculate the numerical simulation model built in step two, and verify the accuracy of the model based on the field data collected in step one. If there are any discrepancies, adjust the parameters.

3. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 1, characterized in that, The Euler two-phase flow model has three phases, specifically: The first phase is a gaseous phase, containing various gaseous components such as: The gaseous components are: O2, CO2, N2, NO, CO, H2O(g), HCN, CH4.C2H4, C2H6, and H2. The second phase consists of pulverized coal solid particles, comprising four components: fixed carbon, volatile matter, ash, and H2O(l). The third phase consists of calcined particles, comprising six components: CaCO3, CAO, C2S, C3S, C3A, and C4AF. Among them, the first phase is the main phase.

4. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 3, characterized in that, The Euler-Euler two-phase flow continuity equation is: In the formula, α q It is the volume fraction of phase q. It is the q-phase velocity. It is the mass flow rate transferred from phase p to phase q. This refers to the mass flow rate transferred from phase q to phase p, S q It is the source phase; The momentum governing equation is: in, In the formula, μ q and λ q q represents the shear viscosity and the bulk viscosity, respectively, F q It is an external force, F lift,q It's lift. It is turbulent dissipation force. Virtual mass force, Rpq, is the interaction force between phases. It is the speed of meeting; The energy control equation is: In the formula, k eff,q It is the thermal conductivity, S q It is the source term, Q pq It is the heat exchange between phase p and phase q, h q It is the enthalpy of the q phase. p op This is the reference atmospheric pressure, and p is the gauge pressure.

5. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 1, characterized in that, in, The pulverized coal combustion simulation model includes a gas combustion model and a solid combustion model; The gas combustion model assumes that the volatile matter in pulverized coal will produce combustible gases; this part uses a fast chemical reaction model. Among them, R i,r It is the reaction rate (kmol / s), v i " ,r With v i ' ,r ε is the stoichiometric coefficient, A and B are empirical constants, 4.0 and 0.5 respectively, YR is the mass fraction of reactants, Mw,r is the mole fraction of reactants, YP is the mass fraction of reactants, Mw,p is the mole fraction of reactants, ε is the turbulent dissipation rate, and k is the turbulence intensity. The solid combustion model uses a reduced-core model: Where P o2 It is the partial pressure of oxygen around the particle, k. fm It is the membrane resistance, k am It is the resistance of the ash layer, k rm It is the coefficient of resistance to surface reaction in the core. In the pulverized coal combustion model, gas combustion is added to the main phase, and solid combustion is added to the second phase. Pulverized coal combustion involves the transfer of carbon components from the fixed carbon in the second phase to carbon dioxide components in the first phase. This chemical reaction involving mass transfer between phases uses the isomerization chemical reaction module. The simulation model for multiphase flow transmission and heat transfer includes an interphase transmission model and a heat transfer model. The interphase transmission model uses the Huilin & Gidaspow traction force model, and the interphase transmission model is a gas-solid transmission model. The force exerted on the particle in the air is: In the formula, C d ρ is the drag coefficient. f Let u be the fluid density (kg / m³), u be the fluid velocity (m / s), and v be the fluid velocity. p Particle velocity (m / s); C D =ψ CDErgun +(1-ψ)C DWen-Yu In the formula, C DErgun It is the Ergun drag coefficient, C DWen-Yu ψ is the Wen-Yu drag coefficient, and ψ is the weighting coefficient with a value of (0,1). The heat transfer model uses the GUNN model, specifically: Among them, a clinker calcination simulation model was added to the third phase, resulting in a total of five chemical models. The reaction type and chemical kinetic parameters of each chemical reaction model were obtained through experiments. The particle enlargement model is as follows: The particle enlargement model uses the PBM model, and the PBM model is solved using the discretization method. The PBM model is coupled with the CFD model to simulate the clinker calcination process. The particle enlargement model is added to the third phase. The Hillert model was used to calculate and predict the clinker calcination model, specifically: Where d0 is the initial particle size, A is the pre-exponential factor for particle growth, and E... p R is the activation energy for particle growth, T is the gas constant, T is the particle temperature, and time is the calcination time. The specific simulation model for pollutant generation is as follows: The pollutant model uses a finite chemical reaction rate model, where the chemical reaction rate is: Among them, C j,r It is the molar concentration (kmol / s), η' i,r With η” i,r It represents the stoichiometric coefficients of reactants and products, Γ represents the effect of reversible reaction, and v″ i,r With v′ i,r It is a stoichiometric coefficient; The chemical reaction rate constant is expressed using the Arrhenius equation: Among them, A r It refers to the pre-factor, where T is temperature, β is the temperature coefficient, and E is the pre-factor. r It is the activation energy, and R is the gas constant; In addition, there are catalytic reduction reactions, which can be modeled by adding a catalyst mass phase to the finite chemical rate model, i.e.: Since the pollutant model components are all gases, they are added to the first phase. At this point, the 3D numerical simulation of the rotary kiln is complete.

6. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 1, characterized in that, After calculating the first 500 seconds of the rotary kiln simulation using the numerical simulation model built in step two, the velocity field, temperature field, and three-phase gas-solid component field data of the entire rotary kiln are exported. By utilizing a flow field feature analysis method based on long short-term memory neural networks and employing an unsupervised training method to fully mine the hidden features in the time history signal, a low-dimensional representation and feature analysis of the complex time-series characteristics of the flow field inside the rotary kiln is performed, resulting in an AI model for rotary kiln operation. The AI ​​model is then used to perform calculations for the subsequent 1300 seconds.

7. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 1, characterized in that, The specific process for low-dimensional characterization and feature analysis of the complex temporal characteristics of the flow field inside the rotary kiln is as follows: First, perform eigenorthogonal decomposition on the sample data to standardize the simulated samples, i.e., the flow field data; let the original data be x. i Where i = 1, 2, 3, ..., r, the sample data x i Let r be an n-dimensional vector, and r be the number of samples; The covariance matrix of the standardized data is as follows: By solving for the eigenvalues ​​of this n×n covariance matrix, its first m eigenvalues ​​can be denoted as λ. 1 , λ 2 、…、λ m The corresponding fundamental mode eigenvectors can be denoted as ξ. 1 ξ 2 、…、ξ m The value of m is determined based on the proportion of the variance of different basic modes to the total variance, ensuring that the feature components contained in the basic modes account for more than 95% of the entire sample space. Therefore, the original sample data can be approximated as: X=U λ ξ; Secondly, feature extraction of time history signals is performed by combining one-dimensional convolution. All convolution kernels in the model are relatively large kernels of size nine, and the number of convolution kernels is thirty-two. The activation function is the ReLU function to obtain local time history information. Long Short-Term Memory (LSTM) neural networks learn long-term dependent information, process and predict important events with relatively long intervals and delays in time series, and avoid the gradient vanishing or gradient exploding problems encountered during backpropagation. LSM neural networks consist of a forget gate, an input gate, and an output gate.

8. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 1, characterized in that, The forgetting gate takes the output ht-1 from the previous layer and the sequence data xt to be input to the current layer as input, passes it through an activation function sigmoid, and obtains the output ft. The output value of ft is in the interval [0,1], representing the probability that the cell state of the previous layer is forgotten, where 1 means complete retention and 0 means complete forgetting. The specific formula is as follows: f t =σ(W f ·[h t-1 ,x t ]+b f ); Output gate: The input gate consists of two parts. The first part uses the sigmoid activation function and outputs "it". The second part uses the tanh activation function and outputs "it". This is the output of this layer; `it` takes a value in the range [0,1]. The degree to which information is preserved in the representation. This indicates that new information is being retained at this layer; i t =σ(W f ·[h t-1 ,x t ]+b i ) So far, ft is the output of the forget gate, which controls the degree to which the previous cell state Ct-1 is forgotten. The two output multiplication operations of the input gate represent how much new information is retained. Based on this, we can update the cell state Ct of this layer with the new information. Output gate: The output gate is used to control the number of filter states for the cell state in this layer; first, the sigmoid activation function is used to obtain an ot value in the range [0,1], then the cell state Ct is processed by the tanh activation function and multiplied by ot, which is the output ht of this layer; the t =σ(W o [h t-1 ,x t ]+b o ) h t = no t ·fish(C) t )。 9. The numerical simulation method for multiple physical phenomena of rotary kilns based on artificial intelligence reasoning according to claim 1, characterized in that, The specific method for learning long-term dependent information using long short-term memory neural networks is as follows: Step 1: Calculate the output value of each neuron in the forward pass; for LSTM, calculate the output value separately according to the algorithm described above; Step 2: Determine the objective function for optimization; In the early stages of training, the output value and the expected value will be inconsistent, so the error term value of each neuron is calculated and the loss function is constructed; Step 3: Update the network weight parameters according to the gradient guidance of the loss function; the backpropagation of the LSTM error term includes two levels: one is the spatial level, which propagates the error term to the previous layer of the network; the other is the temporal level, which propagates backward along time, that is, starting from the current time t, calculates the error at each time step. Then jump to step one, and repeat steps one, two and three until the network error is less than the given value.

Images