Method for predicting NOx concentration at pellet denitration inlet of grate-rotary kiln
By constructing a rotary kiln numerical simulation model and an LSTM neural network, the problems of lagging NOx concentration measurement and poor adaptability to operating conditions in the chain grate-rotary kiln pellet production process were solved, achieving accurate prediction and stable operation of the SCR denitrification system, reducing operating costs, and promoting green and sustainable development.
Patent Information
- Application Number
- CN202511614356.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-06
- Publication Date
- 2026-02-27
AI Technical Summary
In the existing chain grate-rotary kiln pellet production process, the inlet NOx concentration measurement of the SCR denitrification system is lagging and lacks accuracy, has poor adaptability to operating condition fluctuations, and is out of sync with the actual operating conditions, resulting in inaccurate ammonia injection control and problems of excessive or insufficient ammonia injection.
A numerical simulation model of a rotary kiln was constructed, and combined with a component transport model, an LSTM neural network was used to fuse real-time operating data to accurately predict the NOx concentration at the inlet of the SCR denitrification system. The combustion and NOx generation processes were described by the Chemkin PSR model, the GRI-mche3.0 chemical reaction mechanism, and the Realizable k-ε turbulence model. The Navier-Stokes and Brinkman equations were used to simulate NOx transport, and the LSTM network captured the temporal variation patterns.
It achieves real-time and accurate NOx concentration prediction, improves the stability of SCR denitrification systems, reduces operating costs, reduces ineffective ammonia consumption and secondary pollution, meets environmental protection policy requirements, and promotes the green and sustainable development of the pelleting industry.
Smart Images
Figure CN121583369A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of iron ore pellet production technology, and in particular to a method for predicting the inlet NOx concentration of an embedded SCR denitrification system in the chain grate-rotary kiln pellet production process. Background Technology
[0002] The chain grate-rotary kiln process is the mainstream technology for iron ore pellet production worldwide, accounting for over 60% of total iron ore pellet output. This process generates a large amount of NO during the high-temperature roasting stage in the rotary kiln. x NO formation is influenced by a combination of factors, including fuel quality (such as the nitrogen content of coal and natural gas), combustion temperature (peak kiln temperature can reach 1200-1400℃), raw material nitrogen content, and ventilation conditions, leading to NO formation. x Emissions fluctuate significantly.
[0003] To meet environmental protection requirements for ultra-low emissions, the industry widely adopts embedded SCR denitrification technology. This technology utilizes the residual temperature window of 320℃-420℃ in rotary kiln flue gas to react ammonia-based reducing agents with NO under the action of a catalyst. x The reaction produces harmless nitrogen and water, offering the advantages of high denitrification efficiency and low energy consumption. However, existing technologies have the following key problems: Measurement lag and insufficient accuracy: NO at the inlet of the SCR denitrification system x The concentration relies on online analyzer detection, but the transmission delay of flue gas in the pipeline and the analyzer response time mean that the detected value cannot reflect the NO concentration under the current operating conditions in real time. x Actual concentration; at the same time, uneven distribution of flue gas can easily lead to poor representativeness of sampling points, further reducing detection accuracy.
[0004] Poor adaptability to operating condition fluctuations: During the production process of the chain grate rotary kiln, parameters such as pulverized coal injection rate, combustion air flow rate, and secondary air temperature often fluctuate due to changes in raw material batches and output adjustments. Traditional experience-based ammonia injection control cannot timely match NO levels. x Fluctuations in ammonia concentration can easily lead to excessive ammonia injection, causing ammonia escape, secondary pollution, and equipment corrosion; or insufficient ammonia injection can lead to NO₂. x The problem of excessive emissions.
[0005] Numerical simulations are out of sync with actual operating conditions: Existing rotary kiln NO x The generated simulations often neglect the transmission losses of flue gas in the fume hood, wind box, and pellet bed, or fail to consider the impact of cross-flow in the preheating section on NO. x The concentration affects the simulation results, causing a large deviation between the simulation results and the actual inlet concentration, making it difficult to use directly for regulatory guidance.
[0006] Therefore, it is necessary to develop a method that can integrate real-time operating data and accurately simulate NO. xThe generation and transmission process, and the method of achieving time-series prediction, are urgently needed to improve the stability of SCR denitrification systems, reduce operating costs, and promote the green and sustainable development of the pellet industry. Summary of the Invention
[0007] Addressing the NOx issues faced by existing chain grate-rotary kiln pelletizing embedded denitrification systems x To address issues such as inaccurate measurement and delayed detection of NO generation, this invention proposes an inlet NO... x Concentration prediction method. This method accurately calculates NO concentration by constructing a numerical simulation model of a rotary kiln. x The amount of NO generated is calculated, and its migration process in the system is simulated using a component transport model. Finally, an LSTM neural network is used to fuse real-time operating data with simulation results to achieve control of NO at the inlet of the SCR denitrification system. x Accurate prediction of concentration.
[0008] To achieve the above objectives, the present invention provides a chain grate rotary kiln pellet denitrification inlet NO. x Concentration prediction methods include the following steps: S1: Read the production process parameters of the chain grate machine-rotary kiln pelletizing process. These parameters include the rotary kiln main burner combustion air flow rate, pulverized coal injection rate, secondary air flow rate, secondary air temperature, chain grate machine preheating stage hood temperature, wind box exhaust gas temperature, SCR denitrification system inlet temperature, and NO... x concentration; S2: Using the production process parameters described in step S1 as boundary and / or initial conditions, establish a numerical simulation model of the rotary kiln to analyze fuel combustion and NO in the kiln. x The reaction process was simulated to calculate the NO content in the rotary kiln under the current operating conditions. x Production volume; S3: Kiln NO obtained from step S2 x The amount of NO generated was determined using a component transport model. x The transmission process of the rotary kiln, the preheating stage hood of the chain grate machine, and the preheating stage air box was simulated, and the NO at the SCR denitrification inlet was calculated. x Simulated concentration values; S4: Construct NO based on multiple sets of training data x The concentration neural network prediction model includes a set of production process parameters and the corresponding NO values calculated based on the rotary kiln numerical simulation model and the component transport model for each set of training data. x Concentration simulation values, the neural network model is used to calculate based on production process parameters and NO. x The simulated concentration output predicts NO x concentration.
[0009] Further, the numerical simulation model of the rotary kiln in step S2, the construction steps thereof include: S2-1: a 1:1 geometric model of the rotary kiln and the matched pipeline is established, and the geometric model is meshed, and the boundary conditions are determined through the production process parameters read in step S1; S2-2: the Chemkin PSR model, the GRI-mche3.0 chemical reaction mechanism, the Realizable k-ε turbulence model, the radiation heat transfer model, the NO x generation model and the component transport model are used to respectively describe the combustion, the flow, the diffusion and the NO x generation; S2-3: the Zeldovich extension mechanism is used to simulate the generation of thermal NO x , and the generation paths are as shown in formula 1 to formula 4: O2→2O (1) O+N2→NO+N (2) N+O2→NO+O (3) N+OH→NO+H (4) Meanwhile, the concentrations of O free radicals and OH free radicals are calculated according to the partial equilibrium method; S2-4: the kinetic equations of the above elementary reactions are established, and the Arrhenius equation is used to calculate the reaction rate constant of each elementary reaction for NO x generation.
[0010] Further, in step S2-4, for the irreversible reaction, the standard Arrhenius equation is used to calculate the chemical reaction rate : (5) wherein, is the reaction rate constant, and the unit depends on the order of the reaction; is the pre-exponential factor, and the unit depends on the order of the reaction; is the activation energy of the reaction, J / mol; is the gas constant, J / (mol·K); is the reaction temperature, K.
[0011] Further, for the reversible reaction, the forward reaction rate and the reverse reaction rate are calculated, and the reverse reaction rate is calculated through the equilibrium constant : (6) (7) (8) wherein, is the Gibbs free energy change of the reaction, in J / mol; is the standard reaction enthalpy, is the standard reaction entropy; is the reaction temperature, in K.
[0012] NO x The transport process simulation step includes: S3-1: for NO x The diffusion and transport process in the pipe, the flow is described by Navier-Stokes equation, the chemical component mass conservation equation of incompressible fluid is: (9) wherein is the mass fraction of the component, is the chemical reaction rate of the component, in kg·m -3 ·s -1 , is the diffusion coefficient of the component, in m 2 ·s -1 , NO x diffusion in the pipe is calculated by Fuller equation; S3-2: for NO x The diffusion and transport process in the ball material layer, the flow is described by Brinkman equation, the diffusion of NO x in the ball material layer is equivalent to the diffusion process in the porous medium, and the diffusion coefficient is calculated by Knudson formula .
[0013] Further, the Fuller equation in step S3-1 is: (10) wherein, is the temperature, in K; is the pressure, in Pa; refers to the volume diffusion coefficient, m 3 / mol; refers to the amount of molecules, kg / mol.
[0014] Further, the Knudson formula in step S3-2 is: (11) wherein is the temperature, in K, The amount of molecules, measured in kg / mol. It refers to the pore size of the porous medium, measured in meters (m). It is porosity. This is the tortuosity coefficient.
[0015] Furthermore, in step S4, NO x The steps for constructing a concentration neural network prediction model include: S4-1: The model inputs are determined as follows: main burner combustion air flow rate, pulverized coal injection rate, secondary air flow rate, secondary air temperature, chain grate preheating stage hood temperature, and exhaust gas temperature from the wind box. Based on these parameters, the NO at the SCR denitrification inlet is calculated using the rotary kiln numerical simulation model and the transmission process simulation. x Simulated concentration values, and the corresponding NO values for this set of parameters. x Measured concentration value; S4-2: The neural network model is a Long Short-Term Memory (LSTM) network model. The input of the basic LSTM unit includes real-time detection parameters such as the combustion air flow rate and pulverized coal injection rate of the rotary kiln main burner. The cell state is NO from the previous time step. x Measured value, hidden state is NO at the SCR denitrification inlet x Simulated concentration values; S4-3: Design the objective function, optimize the objective function to obtain the optimal LSTM parameters, and substitute the optimal parameters into the LSTM network for prediction.
[0016] The feature is that the neural network model is a Long Short-Term Memory (LSTM) network model, and the method for obtaining it is as follows: For each basic LSTM unit, the input includes: cell state. Hidden state ,enter Its output is The input of the current LSTM basic unit Real-time monitoring parameters for rotary kiln main burner combustion air flow, pulverized coal injection rate, secondary air flow rate, secondary air temperature, chain grate preheating stage hood temperature, and exhaust gas temperature from the wind box, etc., cell status. NO for the previous moment x Measured value, hidden state To obtain the NO at the SCR denitrification inlet through rotary kiln numerical simulation model and transport process simulation calculation. x Simulated concentration values; for each LSTM basic unit, the calculation relationship between its input and output is as follows: (12) (13) (14) (15) (16) (17) (18) in, For the output of the forget gate, For the output of the output gate, For the output of the information gate, This represents the intermediate cell state of the basic unit of LSTM. The output cell state of the LSTM basic unit. The hidden output state of the LSTM basic unit The output of the LSTM basic unit, The LSTM parameters to be estimated are: For matrix dot product calculation, is the tanh activation function. It is the sigmoid activation function.
[0017] Furthermore, for the NOx concentration LSTM prediction model, the following objective function is designed: (19) in, The total number of samples used to train the LSTM model, For each set of real-time detection parameters, the corresponding future time point NO x Measured concentration value The future time NO calculated based on the LSTM model x After optimizing the objective function in formula (19) to obtain the optimal LSTM parameters, the optimal parameters are substituted into the LSTM network. In the application process, the NO concentration prediction value is calculated based on the real-time detection parameters and the simulation model. x The concentration simulation value can be used to obtain the NO at the SCR denitrification inlet. x Accurate predictions of concentration at future moments.
[0018] The beneficial effects of this invention are: Improving prediction accuracy and solving the measurement lag problem: This invention integrates rotary kiln numerical simulation with LSTM neural networks. The numerical simulation model accurately calculates NO through 1:1 geometric modeling and a multi-physics coupling model. x During the generation and transmission process, the LSTM network captures the temporal variation patterns; the combination of the two enables NO xThe relative error of the concentration prediction value and the measured value is significantly reduced, effectively compensating for the transmission delay and response lag defects of the traditional online analyzer, and ensuring that the prediction result can reflect the actual working condition in real time.
[0019] Enhance the adaptability of the working condition and avoid regulation deviation: The method fully considers the fluctuation of parameters such as coal injection amount and secondary air temperature in the production process, and also takes into account the actual scene factors such as preheating stage and secondary stage wind and pellet material layer diffusion. Through the targeted calculation of Navier-Stokes equation, Brinkman equation and Fuller, Knudson formula in the component transport model, the simulation and prediction can adapt to different raw material batches, yield adjustment and other working condition changes, avoiding the problem of excessive or insufficient ammonia injection in traditional experience ammonia injection regulation.
[0020] Integrate simulation and reality to improve engineering practicability: The existing technology often has the problem of disconnection between numerical simulation and actual working condition. The present application uses real-time production parameters as the boundary conditions of the simulation model, and then uses the simulation results and real-time parameters to train the neural network together, realizing a closed loop of physical simulation-data fusion-intelligent prediction. The prediction model not only has physical theory support, but also can adapt to the actual running state of the scene, and the output result can be directly used for ammonia injection amount regulation of the SCR denitration system, with high engineering application value.
[0021] Reduce operating costs and promote green development: High-precision NO x concentration prediction can realize accurate matching of reducing agent and NO x , reduce the invalid consumption of ammonia, and reduce the cost of enterprise reagents; at the same time, avoid the secondary pollution and equipment corrosion caused by ammonia escape, reduce the environmental protection treatment cost and equipment maintenance cost. In addition, stable ultra-low emission meets the requirements of environmental protection policy, helps the chain grate-rotary kiln pellet process to realize green and sustainable development, and improves the environmental protection competitiveness of enterprises. BRIEF DESCRIPTION OF DRAWINGS
[0022] Figure 1 The present application provides a chain grate-rotary kiln pellet production process embedded denitration system inlet NO x concentration prediction method scheme diagram; Figure 2 The present application provides a chain grate-rotary kiln pellet production process embedded denitration system inlet NO x prediction model schematic diagram; Figure 3 The present application provides a chain grate-rotary kiln pellet production process embedded denitration system inlet NO DETAILED DESCRIPTION
[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.
[0024] It should be noted that if the embodiments of the present application involve directionality indication (such as up, down, left, right, front, back, etc.), the directionality indication is only used to explain the relative position relationship, movement condition, etc. between components in a certain posture, and if the certain posture changes, the directionality indication also changes accordingly.
[0025] In addition, if the embodiments of the present application involve descriptions such as "first", "second", etc., the descriptions of "first", "second", etc. are only for description purposes, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first" and "second" can explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing throughout the text includes three parallel solutions. Taking "A and / or B" as an example, it includes A solution, or B solution, or A and B solutions. In addition, "multiple" means two or more. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the fact that a person of ordinary skill in the art can realize it. When the combination of technical solutions contradicts each other or cannot be realized, it should be considered that the combination of technical solutions does not exist and is not within the protection scope of the present application.
[0026] The present application provides a chain grate-rotary kiln pellet denitration inlet NO x The concentration prediction method comprises the following steps: S1-1: reading the chain grate-rotary kiln pellet production process parameters, the production process parameters comprising: rotary kiln main burner combustion-supporting air flow, coal injection amount, secondary air flow, secondary air temperature, chain grate preheating second-stage smoke hood temperature, air bellow exhaust gas temperature, SCR denitration system inlet temperature and NO x concentration, etc. real-time detection parameters; S1-2: based on the rotary kiln numerical simulation model, the fuel combustion process in the rotary kiln is simulated with the real-time detection parameters as initial / boundary conditions, and the NO x generation amount in the rotary kiln under the current working condition is calculated; S1-3: based on the component transport model, considering the NO xThe influence of the transport process, the current working condition of the SCR denitration system inlet NO x concentration is calculated; S1-4: According to a plurality of training data, a NO x concentration neural network prediction model is constructed, and the NO x concentration simulation value outputs the SCR denitration system inlet NO x concentration at the future time. x
[0027] The rotary kiln numerical simulation model described in step S1-2, the construction steps of which include: S2-1: Based on the modeling software, a 1:1 rotary kiln and supporting pipeline geometric model is established, the geometric model is meshed, and the boundary conditions are determined through the parameters read in step S1-1; S2-2: A suitable mathematical model is used to describe combustion, flow, diffusion and NO x generation, specifically including: Chemkin PSR model, GRI-mche3.0 chemical reaction mechanism, Realizable k-ε turbulence model, radiation heat transfer model, NO x generation model, component transport model; S2-3 determines the NO x generation path, the source of NO x is mainly thermal NO x , and the generation mechanism is Zeldovich expansion mechanism. Since the generation of NO x needs to calculate the concentrations of O free radicals and OH free radicals, a partial equilibrium method is used to calculate the concentrations of O free radicals and OH free radicals. The main elementary reactions include: O2→2O (1) O+N2→NO+N (2) N+O2→NO+O (3) N+OH→NO+H (4) S2-4: The kinetic equations of the above elementary reactions are established, and the NO x generation adopts Arrhenius equation to calculate the reaction rate constant of each elementary reaction. For the combustion process of natural gas and coke oven gas, the chemical reaction rate equation can calculate the rate equation of different reactions according to different reactions; (1) For irreversible reactions, the standard Arrhenius equation is used to calculate the chemical reaction rate : (5) Where, is the reaction rate constant, and the unit depends on the order of the reaction; is the pre-exponential factor, and the unit depends on the order of the reaction; is the activation energy of the reaction, J / mol; is the gas constant, J / (mol·K); is the reaction temperature, K; (II) For reversible reactions, not only the forward reaction rate needs to be calculated , but also the reverse reaction rate needs to be calculated. The reverse reaction rate is calculated by the equilibrium constant ; (6) where, is the Gibbs free energy change of the reaction, J / mol; it can be calculated by the standard reaction enthalpy and entropy: (7) It is calculated by the relationship between and : (8) (III) There are also some reactions involving three-body reactions (some reactions require a third substance, i.e. collision body to promote the reaction) and pressure-dependent reactions (at high or low pressure, the reaction rate of some reactions will change with pressure) etc. In S3-1, the GRI-mche3.0 chemical reaction mechanism contains 53 components and 325 elementary reactions, which not only describes the chemical reactions in detail, but also provides optimized reaction rates and reaction paths, reduces the complexity of calculation, and can effectively simulate the species change, free radical reaction, temperature and pressure distribution in the combustion process, etc. It is especially suitable for hydrocarbon combustion process, and the description of free radical reaction in the combustion process also lays a foundation for the subsequent accurate prediction of NO x , in which the main reaction path of carbon element is: , and the main reaction path of hydrogen element is: ; S3-2 uses Chemkin PSR model to describe the chemical reaction process, which has the advantages of relatively simple mathematical model, small amount of calculation, suitable for fast simulation, etc. while ensuring high accuracy, to a certain extent, solves the problem of long time-consuming in batch calculation, etc. The mass, energy and chemical reaction rate equations are described by the following formulas respectively; (9) where, is the concentration of reactants or products, mol·m -3 ; is the mass flow rate, kg·s -1 ; is the reactor volume, m 3; It is the feed concentration, mol·m -3 ; It is the chemical reaction rate, mol·m -3 ·s -1 ; (10) in, It is the temperature inside the reactor, in K; It refers to the input or output of heat, W (J·S). -1 ); It is the feed temperature, in K; (11) in, It is a temperature-dependent rate constant; It is the reactant concentration, mol·m -3 ; It is the stoichiometric coefficient of the reactants, dimensionless; The NO mentioned in steps S1-3 x The conveying process simulation model, combined with the dimensions of the preheating stage 2 and preheating stage 1 of the on-site chain grate machine and the dimensions of the retaining wall, takes into account the air transmission process of the preheating stage 2 and preheating stage 1.
[0028] Step S4-1 for NO x In pipelines, diffusion and transport processes are described by the Navier-Stokes equations. This component model can describe the transport and diffusion phenomena of different components. For incompressible fluids, the mass conservation equations for chemical components are: (12) in, It is the mass fraction of the component, dimensionless; No. Chemical reaction rate of the components, kg·m -3 ·s -1 ; It is the first The diffusion coefficient of each component, m 2 ·s -1 ; For NO x Diffusion in the pipes The calculation is NO x The diffusion coefficient in the gas phase is calculated using the Fuller equation, with the specific formula as follows: (13) in, It is temperature, in K; It is pressure, Pa; refers to the volume diffusion coefficient, m 3 / mol; refers to the amount of molecules, kg / mol; Step S4-2 for NO x In the diffusion and transport process in the pellet layer, the flow is described by using the Brinkman equation, in addition, the NO x diffusion in the pellet layer is equivalent to the diffusion process in the porous medium, and the Knudson formula is used to calculate the diffusion coefficient The specific calculation formula is: (14) Wherein, is the temperature, K; refers to the amount of molecules, kg / mol; is the pore size of the porous medium, m; is the porosity, dimensionless; is the tortuosity coefficient, dimensionless; Step S5-1 NO x The input of the concentration neural network prediction model includes real-time detection parameters such as rotary kiln main burner combustion supporting air flow, coal injection amount, secondary air flow, secondary air temperature, chain grate preheating second stage smoke cover temperature, and air bellow exhaust gas temperature, based on the group of real-time detection parameters, the NO x concentration simulation value at the SCR denitration inlet is calculated by the rotary kiln numerical simulation model and the transmission process simulation, and the NO x concentration measured value corresponding to the group of real-time detection parameters; Step S5-2 the neural network model is a long short-term memory (Long short-term memory, LSTM) model, and the acquisition method is as follows: For each LSTM basic unit, the input includes: cell state , hidden state , input , and the output is . The input of the current LSTM basic unit is the real-time detection parameters such as rotary kiln main burner combustion supporting air flow, coal injection amount, secondary air flow, secondary air temperature, chain grate preheating second stage smoke cover temperature, and air bellow exhaust gas temperature, the cell state is the NO x measured value at the last moment, and the hidden state is the NO x concentration simulation value at the SCR denitration inlet calculated by the rotary kiln numerical simulation model and the transmission process simulation; for each LSTM basic unit, the calculation relationship between the input and the output is as follows: (12) (13) (14) (15) (16) (17) (18) in, For the output of the forget gate, For the output of the output gate, For the output of the information gate, This represents the intermediate cell state of the basic unit of LSTM. The output cell state of the LSTM basic unit. The hidden output state of the LSTM basic unit The output of the LSTM basic unit, The LSTM parameters to be estimated are: For matrix dot product calculation, is the tanh activation function. It is the sigmoid activation function.
[0029] In step S5-3, for the NOx concentration LSTM prediction model, the following objective function is designed: (19) in, The total number of samples used to train the LSTM model, For each set of real-time detection parameters, the corresponding future time point NO x Measured concentration value The future time NO calculated based on the LSTM model x After optimizing the objective function in formula (19) to obtain the optimal LSTM parameters, the optimal parameters are substituted into the LSTM network. In the application process, the NO concentration prediction value is calculated based on the real-time detection parameters and the simulation model. x The concentration simulation value can be used to obtain the NO at the SCR denitrification inlet. x Accurate predictions of concentration at future moments.
[0030] Example 1: The following is in conjunction with the appendix Figures 1-3 Specific embodiments of the present invention are described below: like Figure 1 As shown in this embodiment, the inlet NO of an embedded denitrification system in a chain grate-rotary kiln pellet production process is... xThe concentration prediction method comprises a data processing component and a numerical simulation calculation component, wherein the data processing component is mainly used for extracting boundary conditions (such as rotary kiln primary air parameters, SCR inlet temperature, etc.) required for numerical calculation from the system, and verifying the consistency of the numerical simulation results with the actual data; the numerical simulation calculation component is mainly based on the geometric model of the rotary kiln and the matching pipeline constructed in a 1:1 manner, combined with the boundary conditions obtained by the data processing component, and the appropriate mathematical model is selected to describe the generation and transport process of NO x , and at the same time, the accuracy of the established numerical simulation model is verified, finally, through real-time acquisition of the data (such as secondary air flow, fuel gas flow) in the production process, the above numerical simulation model is substituted, so as to obtain the NO x concentration distribution at the inlet of the embedded denitration system in the chain grate-rotary kiln pellet production process.
[0031] Based on the above numerical simulation-based rotary kiln-SCR inlet NO x prediction method, the embodiment also proposes a chain grate-rotary kiln pellet production process embedded denitration system inlet NO x concentration prediction method, comprising the following steps: I. Reading chain grate-rotary kiln pellet production process parameters mainly including: rotary kiln primary air velocity, air temperature, secondary air temperature and fuel gas flow (natural gas / coke oven gas flow) and other parameters, SCR denitration system inlet temperature and NO x concentration measured value and other parameters; the above parameters are collected through the production line DCS system and online detection equipment, as the boundary conditions for subsequent numerical simulation and the benchmark data for model verification, and at the same time, provide original input data for the LSTM neural network prediction module in the chain grate-rotary kiln pellet production process. Figure 2
[0032] II. Simulating the NO x generation amount in the rotary kiln Based on the combustion model, the combustion process of natural gas / coke oven gas in the rotary kiln is simulated with real-time detection parameters (natural gas / coke oven gas flow, primary / secondary air flow and air temperature, etc.) as boundary conditions, and the NO x generation amount in the rotary kiln under the current working condition is calculated, which specifically includes: Geometric modeling and meshing: based on modeling software such as SolidWorks, the geometric model of the rotary kiln and the matching pipeline is established in a 1:1 manner, the model is meshed through ANSYS ICEM (fine mesh is used in the combustion core area, and regular mesh is used in the pipeline area), and the model boundary conditions are determined by the production parameters read in step I; Mathematical model selection: appropriate mathematical models are used to describe the combustion, flow, diffusion and NO x The generation process specifically includes: Chemkin PSR model (simulates combustion reaction), GRI-mche3.0 chemical reaction mechanism (describes the combustion path of hydrocarbons), Realizable k-ε turbulence model (characterizes turbulent diffusion of gas flow), radiation heat transfer model (calculates heat transfer in the kiln), NO x generation model, component transport model; NO x Generation mechanism determination: clear NO x The main source of NO x is thermal NO x , and its generation mechanism adopts the Zeldovich extended mechanism, and the core reaction path is O2→2O, O+N2→NO+N, N+O2→NO+O, N+OH→NO+H, which ensures that the simulation result is consistent with the NO x generation rule under the high-temperature roasting scene of the rotary kiln.
[0033] III. Simulate the NO x concentration at the inlet of the SCR denitration system Based on the component transport model, the influence of the interflow (air flow movement caused by pressure difference) between the preheating first stage and the preheating second stage of the chain grate machine and the NO x transport process in the pellet layer are mainly considered, and the NO x concentration at the inlet of the SCR denitration system under the current working condition is simulated, specifically including: NO x transport simulation: for the diffusion and transport process of NO x in the smoke cover and the wind box pipeline, the flow state is described by the Navier-Stokes equation, and the diffusion coefficient of NO x in the gas phase is calculated by the Fuller equation, which inputs parameters such as flue gas temperature, pressure, and component molar mass, to accurately calculate the NO x concentration decay in the pipeline; NO x transport simulation in the pellet layer: for the diffusion and transport process of NO x in the pellet layer, the flow state is described by the Brinkman equation, which equates the pellet layer to a porous medium, and the diffusion coefficient is calculated by the Knudson formula (which needs to input parameters such as average pore size, porosity, and tortuosity of the layer), to correct the interception and diffusion influence of the layer on NO x ; Application of simulation results: the simulated value of the NO x concentration at the inlet of the SCR obtained in this step is imported together with the measured value of the NO Figure 2 concentration collected in step one into the simulation database shown, as one of the core training data of the LSTM neural network prediction model.
[0034] IV. LSTM Neural Network Prediction and Result Verification Model training: such as Figure 2 As shown, the production process parameters collected in step one (pellet mass flow, secondary air temperature / volume, gas flow, etc.) and the NO obtained in step three are used. x The concentration simulation value is the input feature, with the future NO collected in step one as the input feature. x The measured concentration value is used as the output label, and an LSTM neural network prediction model is built through offline training. During the training process, the deviation between the predicted value and the actual value of the denitrification inlet is monitored in real time. If the deviation meets the set threshold, the model training is completed and it is used for real-time prediction. Multi-condition verification: Five typical operating conditions of the production line were selected, such as full-load production, raw material batch change, secondary air temperature fluctuation, low-load production, and abnormal exhaust gas temperature in the air box, for predictive verification. The results are as follows: Figure 3 As shown: the horizontal axis represents the operating condition number (operating condition 1-operating condition 5), and the vertical axis represents NO. x Concentration (unit: mg / m³) 3 The curve shows that the deviation between the predicted and measured values is ≤4% under each working condition. For example, the measured value for working condition 1 is 332 mg / m³. 3 Predicted 330 mg / m 3 The measured value under operating condition 5 was 295 mg / m³. 3 Predicted 293 mg / m³ 3 This fully verifies the prediction accuracy of the proposed method; This invention can accurately predict NO at the SCR denitrification inlet using both numerical simulation and LSTM prediction, under changing production conditions. x The concentration provides real-time guidance for the control of ammonia injection volume, thereby achieving precise matching of reducing agent and NOx. This not only improves denitrification efficiency and significantly reduces ammonia slip, but also reduces ammonia consumption and equipment corrosion, which is of great significance for promoting the green transformation and sustainable development of the pelletizing industry.
[0035] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.
Claims
1. A chain grate machine - rotary kiln pellet denitrification inlet NO x Concentration prediction method, characterized in that, Includes the following steps: S1: Read the production process parameters of the chain grate machine-rotary kiln pelletizing process. These parameters include the rotary kiln main burner combustion air flow rate, pulverized coal injection rate, secondary air flow rate, secondary air temperature, chain grate machine preheating stage hood temperature, wind box exhaust gas temperature, SCR denitrification system inlet temperature, and NO... x concentration; S2: Using the production process parameters described in step S1 as boundary and / or initial conditions, establish a numerical simulation model of the rotary kiln to analyze fuel combustion and NO in the kiln. x The reaction process was simulated to calculate the NO content in the rotary kiln under the current operating conditions. x Production volume; S3: Kiln NO obtained from step S2 x The amount of NO generated was determined using a component transport model. x The transmission process of the rotary kiln, the preheating stage hood of the chain grate machine, and the preheating stage air box was simulated, and the NO at the SCR denitrification inlet was calculated. x Simulated concentration values; S4: Construct NO based on multiple sets of training data x The concentration neural network prediction model includes a set of production process parameters and the corresponding NO values calculated based on the rotary kiln numerical simulation model and the component transport model for each set of training data. x Concentration simulation values, the neural network model is used to calculate based on production process parameters and NO. x The simulated concentration output predicts NO x concentration.
2. The method according to claim 1, characterized in that, The construction steps of the rotary kiln numerical simulation model described in step S2 include: S2-1: Establish a 1:1 geometric model of the rotary kiln and its supporting pipelines, and mesh the geometric model. Determine the boundary conditions using the production process parameters read in step S1. S2-2: The Chemkin PSR model, GRI-mche3.0 chemical reaction mechanism, Realizable k-ε turbulence model, radiative heat transfer model, and NO were employed. x The generation model and component transport model describe combustion, flow, diffusion, and NO, respectively. x generate; S2-3: Simulation of thermal NO using the Zeldovich extended mechanism x The generation path is shown in Equations 1 to 4: O2→2O (1) O+N2→NO+N (2) N+O2→NO+O (3) N + OH → NO + H (4) Meanwhile, the concentrations of O radicals and OH radicals were calculated using a partial equilibrium method; S2-4: Establish the kinetic equations for the above elementary reactions, specifically for NO. x The reaction rate constants for each elementary reaction are generated using the Arrhenius equation.
3. The prediction method according to claim 1, characterized in that, In steps S2-4, for irreversible reactions, the chemical reaction rate is calculated using the standard Arrhenius equation. : （5） in, It is the reaction rate constant, and its unit depends on the order of the reaction; It refers to the pre-factor, and the unit depends on the order of the reaction; It is the activation energy of the reaction, in J / mol; It is the gas constant, and its unit is J / (mol·K); It is the reaction temperature, measured in K.
4. The prediction method according to claim 1, characterized in that, For a reversible reaction, calculate the forward reaction rate. and reverse reaction rate The reverse reaction rate is related to the equilibrium constant. calculate: （6） （7） （8） in, It is the Gibbs free energy change of the reaction, with units of J / mol; It is the standard enthalpy of reaction. It is the standard reaction entropy; It is the reaction temperature, measured in K.
5. The method according to claim 1, characterized in that, In step S3, NO x The simulation steps of the transmission process include: S3-1: For NO x The diffusion and transport processes in pipes are described by the Navier-Stokes equations, and the mass conservation equations for the chemical components of incompressible fluids are: （9） in It is the mass fraction of the component. It is the first Chemical reaction rate of each component, in kg·m -3 ·s -1 , It is the first The diffusion coefficient of each component, in m 2 ·s -1 NO x Diffusion in the pipe Calculated using Fuller's equations; S3-2: For NO x The diffusion and transport processes within the spherical material layer are described by the Brinkman equation, which describes the NO flow. x Diffusion in the pellet bed is equivalent to diffusion in a porous medium. The diffusion coefficient is calculated using the Knudson formula. .
6. The method according to claim 5, characterized in that, The Fuller equation in step S3-1 is: （10） in, It's temperature, and the unit is K. It is pressure, and the unit is Pa; This refers to the volume diffusion coefficient, m 3 / mol; This refers to the amount of molecules, expressed in kg / mol.
7. The method according to claim 5, characterized in that, The Knudson formula in step S3-2 is: （11） in It is temperature, measured in Kelvin (K). The amount of molecules, measured in kg / mol. It refers to the pore size of the porous medium, measured in meters (m). It is porosity. This is the tortuosity coefficient.
8. The method according to claim 1, characterized in that, In step S4, NO x The steps for constructing a concentration neural network prediction model include: S4-1: The model inputs are determined as follows: main burner combustion air flow rate, pulverized coal injection rate, secondary air flow rate, secondary air temperature, chain grate preheating stage hood temperature, and exhaust gas temperature from the wind box. Based on these parameters, the NO at the SCR denitrification inlet is calculated using the rotary kiln numerical simulation model and the transmission process simulation. x Simulated concentration values, and the corresponding NO values for this set of parameters. x Measured concentration value; S4-2: The neural network model is a Long Short-Term Memory (LSTM) network model. The input of the basic LSTM unit includes real-time detection parameters such as the combustion air flow rate and pulverized coal injection rate of the rotary kiln main burner. The cell state is NO from the previous time step. x Measured value, hidden state is NO at the SCR denitrification inlet x Simulated concentration values; S4-3: Design the objective function, optimize the objective function to obtain the optimal LSTM parameters, and substitute the optimal parameters into the LSTM network for prediction.
9. The method according to claim 8, characterized in that, The neural network model is a Long Short-Term Memory (LSTM) network model, and the method for obtaining it is as follows: For each basic LSTM unit, the input includes: cell state. Hidden state ,enter Its output is The input of the current LSTM basic unit Real-time monitoring parameters for rotary kiln main burner combustion air flow, pulverized coal injection rate, secondary air flow rate, secondary air temperature, chain grate preheating stage hood temperature, and exhaust gas temperature from the wind box, etc., cell status. NO for the previous moment x Measured value, hidden state To obtain the NO at the SCR denitrification inlet through rotary kiln numerical simulation model and transport process simulation calculation. x Simulated concentration values; for each LSTM basic unit, the calculation relationship between its input and output is as follows: （12） （13） （14） （15） （16） （17） （18） in, For the output of the forget gate, For the output of the output gate, For the output of the information gate, This represents the intermediate cell state of the basic unit of LSTM. The output cell state of the LSTM basic unit. The hidden output state of the LSTM basic unit The output of the LSTM basic unit, The LSTM parameters to be estimated are: For matrix dot product calculation, is the tanh activation function. It is the sigmoid activation function.
10. The method according to claim 8, characterized in that, For the LSTM prediction model of NOx concentration, the following objective function is designed: （19） in, The total number of samples used to train the LSTM model, For each set of real-time detection parameters, the corresponding future time point NO x Measured concentration value The future time NO calculated based on the LSTM model x After optimizing the objective function in formula (19) to obtain the optimal LSTM parameters, the optimal parameters are substituted into the LSTM network. In the application process, the NO concentration prediction value is calculated based on the real-time detection parameters and the simulation model. x The concentration simulation value can be used to obtain the NO at the SCR denitrification inlet. x Accurate predictions of concentration at future moments.