SCR denitration reactor optimization method based on CFD flow field simulation

Abstract

The invention relates to the technical field of flue gas pollutant control, and discloses an SCR denitration reactor optimization method based on CFD flow field simulation, and the method comprises the following steps: S1, building an initial CFD model, carrying out reference simulation, and obtaining the performance data of the clean state of a reactor; s2, constructing an ash deposition dynamic evolution model, and performing flow field-ash deposition-geometric coupling simulation in combination with a dynamic grid technology to obtain time-varying operation data; s3, defining and quantifying a time domain performance index of the reactor in a preset period by using the time-varying operation data; and S4, optimizing the design variables by adopting a multi-objective optimization algorithm based on an agent model and taking the time domain performance index as an objective to obtain an optimal design scheme. Through dust deposition dynamic coupling simulation and agent model multi-objective optimization, the long-term operation stability and the comprehensive denitration performance of the SCR reactor under the actual dust deposition condition are remarkably improved, and the optimization design efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of flue gas pollutant control, and in particular to an SCR denitration reactor optimization method based on CFD flow field simulation. Background Art

[0002] Selective catalytic reduction (SCR) denitrification technology, as an efficient and economical method for controlling nitrogen oxides (NOx), has been widely used in industrial sectors such as coal-fired power plants, steel sintering, cement kilns, and waste incineration. However, in practical industrial applications, especially when treating highly dusty flue gases, SCR denitrification reactors often face serious ash accumulation. Fly ash particles carried in the flue gas can deposit, adhere, and gradually accumulate on the guide plates, catalyst surfaces, and other structural components within the reactor. This not only blocks the flue gas flow path and significantly increases system operating resistance, but also degrades the uniformity of the flue gas flow field at the catalyst inlet, affecting the distribution of the NH3 / NOx molar ratio, leading to reduced denitrification efficiency, increased ammonia slip, and even catalyst wear, poisoning, or sintering, severely impacting the long-term stable operation and economic benefits of the SCR system. Therefore, fully considering the dynamic effects of ash accumulation during the design phase and optimizing the design of SCR denitrification reactors to improve their overall performance under actual ash accumulation conditions is of great engineering significance and application value.

[0003] Computational fluid dynamics (CFD) simulation technology is currently widely used in the design and optimization of SCR denitrification reactors to analyze the flow, temperature, and species concentration distributions within the reactor, thereby guiding structural improvements. Designers typically construct a three-dimensional geometric model of the reactor and perform CFD simulations based on ideal, clean, and dust-free operating conditions to evaluate initial denitrification efficiency, flow uniformity, and pressure drop. In some cases, to account for the effects of dust accumulation, simplified approaches may be employed, such as pre-setting a simulated dust layer of fixed thickness on the wall or estimating the impact of dust accumulation on pressure drop using empirical formulas. Regarding optimization, some studies have attempted to optimize certain reactor structural parameters (such as guide vane angle and position) based on CFD results for a single or limited number of performance indicators through parametric studies or incorporating optimization algorithms (such as genetic algorithms and particle swarm optimization).

[0004] While existing technologies have made some progress in the design and understanding of SCR denitrification reactors, several shortcomings remain. First, traditional CFD simulations are mostly based on an initial clean state or employ a static, simplified treatment of the effects of soot accumulation. This approach struggles to accurately capture the dynamic temporal and spatial accumulation of soot and its complex coupled effects on flow, temperature, and chemical reactions. This is because soot accumulation is a gradual, dynamic process. The formation of a soot layer in turn alters the reactor's geometric boundaries and internal flow field, influencing the subsequent particle deposition rate and location. Static models cannot fully capture this strong coupling and time-varying nature of the flow field, soot accumulation, and geometry. Consequently, predictions of the reactor's long-term performance under realistic soot accumulation conditions, particularly the decline in denitrification efficiency and the growth of pressure drop, are inaccurate. Second, existing optimization design methods often focus on improving a single performance metric under ideal clean conditions. Even when considering soot accumulation, they struggle to address multiple, dynamically changing and conflicting performance objectives, such as the time-averaged denitrification efficiency over the entire operating cycle, the maximum system pressure drop within the cycle, and the time-averaged uniformity of the catalyst inlet velocity distribution. This is because the computationally expensive simulations that reflect the dynamic process of dust accumulation need to be directly embedded in the multi-objective optimization iteration loop. This computational complexity makes it difficult for designers to systematically explore the vast design space, making it difficult to find a robust design solution that truly achieves a balance of multiple performance aspects in a complex dust accumulation environment. Finally, due to the lack of in-depth understanding and effective prediction methods of the dynamic behavior of dust accumulation and its impact on the overall performance of the reactor, existing designs are often limited in the formulation of anti-dust accumulation and dust management strategies. As a result, in actual operation, they tend to rely more on later, passive dust cleaning measures. This not only increases maintenance costs and operating energy consumption, but also makes it difficult to fundamentally improve the long-term operational stability and economic efficiency of the reactor. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an SCR denitrification reactor optimization method based on CFD flow field simulation, which solves the problem that it is difficult to accurately simulate the dynamic accumulation process of dust accumulation and its complex coupling influence on reactor performance during the design optimization of the SCR denitrification reactor in the existing technology, and it is difficult to efficiently perform multi-objective collaborative optimization while considering this dynamic influence, which makes it difficult to fully guarantee the long-term operating performance and economy of the reactor under actual dust-laden flue gas conditions.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an SCR denitrification reactor optimization method based on CFD flow field simulation, comprising the following steps:

[0007] S1. Establishing an initial computational fluid dynamics model of the SCR denitration reactor, and performing a benchmark CFD simulation on the initial computational fluid dynamics model based on preset boundary conditions and a physical model to obtain performance data of the reactor in an initial clean state;

[0008] S2. Based on the established initial computational fluid dynamics model and the performance data in the initial clean state, a dynamic evolution model of dust accumulation is constructed to describe the change of dust thickness on the inner wall of the reactor over time. The geometry of the initial computational fluid dynamics model is dynamically updated in combination with dynamic mesh deformation technology. A flow field-dust accumulation-geometry dynamic coupling simulation is performed to obtain continuous time-varying operating data of the reactor under the influence of the dust accumulation process.

[0009] S3. Using the continuous time-varying operating data, define and quantify time-domain performance indicators of the SCR denitration reactor within one or more preset operating cycles;

[0010] S4. A multi-objective optimization algorithm based on an agent model is used to perform collaborative optimization with one or more of the quantified time-domain performance indicators as optimization targets and the adjustable design variables of the SCR denitrification reactor as optimization inputs, thereby obtaining an optimized SCR denitrification reactor design scheme.

[0011] The present invention provides an SCR denitrification reactor optimization method based on CFD flow field simulation. It has the following beneficial effects:

[0012] 1. This invention constructs a dynamic evolution model of dust accumulation and combines it with dynamic grid deformation technology to achieve dynamic coupled simulation of the flow field, dust accumulation, and reactor geometry. This allows for more realistic simulation of the continuous, time-varying operation of an SCR denitrification reactor under dust accumulation conditions. This refined and dynamic consideration of the impact of dust accumulation significantly improves the accuracy and reliability of predictions of the reactor's long-term operating performance, avoiding the significant deviations that can occur with traditional static or simplified models when evaluating reactor performance in dusty flue gas environments.

[0013] 2. This invention utilizes a multi-objective optimization algorithm based on a surrogate model, enabling collaborative optimization of the adjustable design variables of the SCR denitrification reactor using multiple quantified time-domain performance indicators as optimization targets. This multi-objective collaborative optimization strategy helps find a better balance between mutually constrained performance requirements, thereby achieving a design solution with better overall performance under the influence of dust accumulation, rather than focusing solely on a single static indicator.

[0014] 3. This invention significantly improves the efficiency of design optimization by introducing a proxy model, replacing the computationally expensive, full flow field-ash accumulation-geometry dynamic coupled simulation with numerous optimization iterations. This allows for the exploration of a wider range of design parameter spaces and more complex ash accumulation management strategies within limited computing resources and timeframes, thereby more effectively obtaining customized, high-performance SCR denitrification reactor optimization solutions tailored to dusty flue gas characteristics and specific operating conditions.

[0015] 4. Through detailed numerical simulation of the dust accumulation process within the reactor and its interaction with the flow field, this invention provides in-depth insight into the formation and development mechanisms of dust accumulation, and its impact on the flow characteristics, heat and mass transfer, and chemical reaction processes within the reactor. This in-depth understanding helps reveal dust-prone areas, key influencing factors, and the specific pathways by which dust accumulation leads to performance degradation, thereby providing a scientific basis for developing more targeted dust prevention and cleaning measures, as well as structural improvement plans.

[0016] 5. By fully considering the dynamic impact of ash accumulation during the design phase and performing multi-objective optimization, this invention helps to fundamentally improve the SCR denitrification reactor's resistance to ash accumulation and long-term operational stability. This effectively reduces problems such as unplanned downtime, increased maintenance frequency, increased energy consumption, and premature catalyst failure caused by ash accumulation, thereby reducing the reactor's full lifecycle operating costs and improving the overall reliability and economic benefits of the device. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the specification of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] Please see the attached Figure 1 The embodiment of the present invention provides an SCR denitrification reactor optimization method based on CFD flow field simulation, comprising the following steps:

[0020] S1. Establish an initial computational fluid dynamics model of the SCR denitrification reactor and perform a benchmark CFD simulation on the initial computational fluid dynamics model based on preset boundary conditions and physical models to obtain performance data of the reactor in the initial clean state;

[0021] The implementation of S1 is:

[0022] First, an initial computational fluid dynamics model was constructed. The purpose of the model is to reproduce the complex fluid flow, heat transfer, and chemical reaction processes inside the SCR denitrification reactor through numerical methods.

[0023] During the model construction process, it is necessary to use professional 3D computer-aided design (CAD) software, such as but not limited to commercial software such as SolidWorks, CATIA, or AutoCAD, to build an accurate 3D geometric model of the SCR denitrification reactor. This 3D geometric model should fully reproduce the actual physical boundaries and internal structural components of the reactor, including the reactor's inlet flue, outlet flue, possible guide plates, the precise dimensions and spatial layout of the catalyst layer area, and auxiliary structures such as ash hoppers. The level of detail of the geometric model should be sufficient to capture key features that have a significant impact on flow characteristics, such as changes in the flue cross-section, the curvature and angle of attack of the guide plates, and the arrangement of the catalyst modules, thereby providing an accurate geometric domain for subsequent numerical simulations.

[0024] After constructing the 3D geometric model, it is necessary to generate a computational mesh. This process discretizes the continuous geometric domain into a finite number of tiny control volume elements, enabling the application of finite volume methods or other numerical methods to solve the governing fluid dynamics equations, such as the Navier-Stokes equations, the energy equation, and the species transport equation. Preferably, a structured, unstructured, or hybrid meshing strategy can be selected based on the complexity of the reactor geometry and the flow characteristics. In areas with sharp gradients in flow field parameters (such as velocity, pressure, and species concentration) or complex geometry (e.g., guide plate edges, sharp bends in the flue, and catalyst inlet interfaces), local mesh refinement is required to improve computational accuracy in these critical areas. Mesh quality is crucial for ensuring convergence and accuracy in CFD simulations. Therefore, after mesh generation, key quality parameters such as mesh orthogonality, aspect ratio, twist, or tilt must be checked and evaluated to ensure they are within acceptable ranges to avoid computational divergence or distorted results due to poor mesh quality.

[0025] Subsequently, the initial computational fluid dynamics model was set with pre-defined physical models, which were used to mathematically describe the complex physical and chemical phenomena occurring inside the reactor. These models were selected based on a deep understanding of the SCR denitrification process mechanism.

[0026] Among them, turbulence models are essential because the flue gas flow in the SCR denitrification reactor is usually in a turbulent state with high Reynolds numbers due to high flow velocity and large geometric scale. Turbulent flow is characterized by irregularity, three-dimensionality, high diffusivity and dissipation.

[0027] Preferably, an appropriate turbulence model can be selected based on the specific flow characteristics and computational resources, such as the Reynolds-averaged Navier-Stokes (RANS) model, such as the standard k-ε model, the RNG k-ε model, and the Realizable k-ε model. Alternatively, for complex flow phenomena such as strong swirl, flow separation, or anisotropic turbulence, a shear stress transport model such as the SST k-ω model can be considered. If computational resources permit, even more sophisticated turbulence models such as the large eddy simulation (LES) or the detached eddy simulation (DES) can be used.

[0028] The selected turbulence model should be able to reasonably predict the average flow characteristics and turbulent mixing effects, which is crucial for subsequent mass transfer and chemical reaction simulations.

[0029] The setting of the chemical reaction model is crucial for accurately simulating the SCR denitrification process. The model needs to include the main NOx reduction reactions, such as the standard SCR reaction between ammonia (NH3) and nitric oxide (NO) (e.g., 4NO+4NH3+O2→4N2+6H2O) and the fast SCR reaction involving nitrogen dioxide (NO2) (e.g., NO+NO2+2NH3→2N2+3H2O). In addition, depending on the actual flue gas composition and catalyst characteristics, side reactions such as NH3 oxidation may need to be considered, which may consume NH3 or produce undesirable by-products. Component transport equations are usually used to describe the convection, diffusion, and source terms (generated or consumed by chemical reactions) of each gas component. The rate of the chemical reaction can be characterized by a volume reaction model (assuming that the reaction occurs uniformly in the fluid domain) or by defining a surface reaction model on the catalyst surface (which is more in line with the actual situation of the catalytic reaction). The reaction rate constant can be obtained by referring to published literature data or by experimental measurement, and the effect of temperature on the reaction rate needs to be considered, such as using the Arrhenius equation ( ) to describe this dependency, where is the pre-exponential factor; is the activation energy; is the chemical reaction rate constant; is the ideal gas constant; is the absolute temperature of the reaction system.

[0030] The discrete phase model (DPM) is introduced to simulate the motion and behavior of fly ash particles carried in flue gas. These fly ash particles are the primary cause of subsequent ash accumulation. Preferably, a Lagrangian approach is used to track the trajectories of a large number of representative particles, solving the particle equations of motion in a Lagrangian coordinate system. This equation typically considers the drag force on the particles from the fluid, gravity, virtual mass force, pressure gradient force, lift forces (such as Saffman lift and Magnus lift), as well as collisions between particles and with walls (possibly accompanied by rebound or capture). Depending on the particle concentration, one can choose between one-way coupling (in which the fluid affects particle motion, but the effect of particles on the fluid is negligible, suitable for sparse particle flows) or two-way coupling (in which momentum, heat, and mass are exchanged between the fluid and particles, suitable for denser particle flows). Particle physical properties, such as particle size distribution (typically conforming to a statistical distribution such as the Rosin-Rammler distribution), density, and shape factor, need to be determined based on the actual fuel characteristics and combustion conditions. The simulation results of DPM, especially the concentration, impact velocity and impact angle of particles near the wall, are the basic data source for the calculation of particle deposition flux in the dynamic evolution model of dust accumulation in the subsequent step S2.

[0031] After completing the physical model setup, it is necessary to set the default boundary conditions for the initial CFD model. These boundary conditions define the interaction between the computational domain and the external environment and are essential for solving the governing equations in partial differential form.

[0032] Inlet boundary conditions require a detailed definition of the flue gas conditions entering the reactor. These include the average flue gas velocity profile (or total mass flow rate), average temperature distribution, operating pressure, and the concentration or mass fraction distribution of various gas components (such as N₂, O₂, H₂O, CO₂, SO₂, as well as key reactants such as NOx and the reducing agent NH₃). In practical applications, inlet parameters may not be uniformly distributed. Therefore, if conditions permit, a more realistic inlet profile can be used. These parameters directly determine the mass and energy flows entering the reaction system.

[0033] Outlet boundary conditions are typically set as pressure outlets, for example, defining the average gauge pressure of the outlet cross-section to be zero (i.e., relative to atmospheric pressure) or to the actual operating backpressure of the downstream equipment. In some cases, an outflow boundary condition can also be used, assuming that the outlet flow is fully developed.

[0034] The setting of wall boundary conditions is very important for simulating the interaction between fluid and solid wall and catalyst. For solid walls such as reactor shells and guide plates, no-slip boundary conditions are usually applied, that is, the fluid velocity at the wall is assumed to be the same as the wall velocity (zero if the wall is stationary). The temperature condition can be set to a constant wall temperature, a constant heat flux, or to consider convective heat transfer with the external environment (defining the heat transfer coefficient and the ambient temperature) according to actual conditions. For the catalyst area, if it is simplified as a porous medium model, its permeability (describing the difficulty of the fluid passing through the porous medium) and inertial resistance coefficient (describing the additional resistance at high speed flow) need to be set to simulate the pressure loss generated by the fluid flowing through the catalyst layer; if the microstructure of the catalyst is explicitly modeled or regarded as part of the fluid domain, the occurrence of the catalytic reaction is defined on its surface or volume, and a specific wall roughness may need to be set.

[0035] After completing the above model construction and boundary condition setting, a baseline CFD simulation can be performed on the initial computational fluid dynamics model. This simulation aims to obtain the internal flow field characteristics and denitrification performance benchmark of the reactor in its initial design state (i.e., clean, dust-free state).

[0036] The established mathematical model including geometry, mesh, physical model and boundary conditions is numerically solved by professional CFD solving software (such as commercial software such as ANSYSFluent, CFX, STAR-CCM+, or open source software such as OpenFOAM). The solution process usually adopts the finite volume method to integrate the control equation on each control volume to obtain a set of algebraic equations, which are then solved by an iterative algorithm. The solution process can be a steady-state simulation to obtain the time-averaged performance of the reactor under stable conditions; or in certain specific cases, such as when it is necessary to analyze flow instability or certain transient operations, a transient simulation can also be performed. During the solution process, it is necessary to monitor the residuals of key variables (indicating the degree of satisfaction of the control equation at each iteration step) and the changes in several physical quantities (such as average velocity, temperature, pressure drop, outlet component concentration, etc.) at selected monitoring points or monitoring surfaces until the preset convergence criteria are reached. The preferred convergence criteria can include the normalized residual of each control equation being less than 10 -4 to 10 -6 The values ​​of key monitoring parameters do not change significantly or fluctuate within the allowable range within multiple consecutive iterations.

[0037] After the baseline CFD simulation is completed, a detailed post-processing analysis of the calculation results is required to extract and quantify the performance data of the reactor in its initial clean state.

[0038] Performance data includes, but is not limited to, velocity field distribution characteristics: By analyzing streamlines, velocity vectors, and velocity isosurfaces / cloud maps within the reactor, one can intuitively determine whether the flue gas flow is smooth and uniform, and whether there are dead zones (low-velocity areas), eddy currents, or severe flow deviations. In particular, the uniformity of the velocity distribution at the catalyst inlet cross section is important, as uneven velocity distribution can lead to reduced catalyst utilization, increased local wear, and ammonia slip.

[0039] Temperature field distribution characteristics: Analyze the temperature distribution inside the reactor and the catalyst area to ensure that the catalyst operates within its optimal activity temperature window (usually SCR catalysts have a specific operating temperature range), while avoiding local overheating (which may lead to catalyst sintering or increased side reactions) or supercooling points (which may lead to reduced catalyst activity or condensation of ammonium bisulfate), which may affect the denitrification efficiency or cause catalyst deactivation.

[0040] Pressure Field Distribution Characteristics: Analyze the static and total pressure distribution within the reactor and calculate the overall pressure drop in the reactor. Pressure drop is a key indicator for evaluating reactor energy consumption. Excessively high pressure drop requires a larger induced draft fan, increasing operating costs. NH3 / NOx Molar Ratio Distribution Characteristics: Detailed analysis of the uniformity of the NH3 to NOx molar ratio (commonly referred to as the ammonia-nitrogen ratio or α value) at the catalyst inlet cross section. An ideal ammonia-nitrogen ratio and its uniform distribution are key to achieving efficient denitrification and controlling ammonia slip. Locally high ammonia-nitrogen ratios can lead to excessive ammonia slip, while locally low ammonia ratios can result in insufficient NOx conversion.

[0041] NOx conversion rate: Calculate the overall NOx removal efficiency of the reactor based on the mass flow rate or average concentration of NOx at the inlet and outlet ( ); where: is the NOx conversion rate or NOx removal efficiency; is the amount of NOx in the flue gas before entering the SCR denitrification reactor; It is the amount of NOx in the flue gas after leaving the SCR denitrification reactor.

[0042] S2. Based on the established initial computational fluid dynamics model and the performance data under the initial clean state, a dynamic evolution model of dust accumulation is constructed to describe the change of dust thickness on the reactor inner wall over time. The geometry of the initial computational fluid dynamics model is dynamically updated by combining dynamic mesh deformation technology. A dynamic coupled flow field-dust accumulation-geometry simulation is performed to obtain continuous time-varying operating data of the reactor under the influence of the dust accumulation process.

[0043] The implementation of S2 is:

[0044] The core of this step is to simulate the inevitable ash accumulation during actual SCR denitrification reactor operation and accurately evaluate the dynamic impact of ash accumulation on the reactor's internal flow field, heat and mass transfer, and chemical reaction performance over time. To achieve this goal, this step innovatively couples the formation and development of ash accumulation, the dynamic changes in the reactor's geometric boundaries, and the CFD simulation of the internal flow field.

[0045] First, a dynamic evolution model of dust accumulation needs to be constructed. This model aims to quantitatively describe the dynamic evolution of any microelement on the reactor wall. Dust thickness at Over time An optimal dynamic evolution model of dust accumulation is formed by calculating the net growth rate of the dust accumulation thickness of the wall microelement, and its mathematical expression is:

[0046] ;

[0047] Where, for Momentary wall element Thickness of dust accumulation; is the effective density of dust accumulation; for Momentary wall element particle mass deposition flux on the surface; for Momentary wall element The mass scouring or stripping flux of the ash deposited on the wall is calculated based on the mass concentration of particles in a specific size range in the local flue gas near the wall obtained in the CFD simulation step of the current flow field-ash deposit-geometry dynamic coupling simulation. , and the particles hitting the wall The average normal velocity component of Both parameters can be obtained by tracking the particle motion trajectory and statistically analyzing the interaction information between the particles and the wall using the discrete phase model (DPM) in step S1. In addition, the calculation of the particle mass deposition flux also requires a preset effective adhesion probability. .

[0048] The adhesion probability is a key empirical or semi-empirical function that characterizes the proportion of particles that hit the wall that can actually adhere and become part of the dust deposit. This probability may not be a constant, but is affected by a combination of factors, such as the thickness of the deposited dust. , the physical and chemical properties of the particles themselves (such as particle size, shape, temperature, viscosity), the wall material properties (such as roughness, temperature, surface energy) and the local flue gas conditions (such as temperature, flow rate, gas composition), etc. These factors are collectively referred to as .

[0049] The main deposition mechanisms of particles may include inertial collision (larger particles), turbulent diffusion (smaller particles), thermophoretic forces (in the presence of temperature gradients), gravitational sedimentation, etc. The relative importance of these mechanisms depends on the particle size and flow conditions.

[0050] Representatives in moment, wall element The mass scouring or stripping flux of dust per unit area on the wall is the mass of dust removed from the unit wall area by fluid dynamics per unit time. The calculation of this flux is mainly based on the wall shear stress obtained in the CFD simulation step of the current flow field-dust-geometry dynamic coupling simulation. , the shear stress is generated when the fluid flows over the dust accumulation surface.

[0051] The peeling of dust accumulation also depends on the bonding strength of the dust accumulation layer itself. , which is a parameter that characterizes the ability of the dust layer to resist external damage. The bonding strength is also a complex function, which may be related to the dust thickness. , the "age" of dust accumulation The bond strength of the ash layer is affected by several factors, including the time the ash has been deposited, during which aging processes such as sintering and phase transformation may occur, leading to changes in bond strength. The physical and chemical composition and microstructure of the ash deposit are also factors. When the wall shear stress generated by flue gas flow exceeds the critical bond strength of the ash layer, the ash will be washed away or separated into chunks.

[0052] Secondly, in order to achieve dynamic simulation of the effect of dust accumulation on the geometric shape of the reactor, the present invention adopts dynamic mesh deformation technology. As dust accumulates on the inner wall of the reactor, the actual geometric boundary of the flow channel will change, thereby affecting the internal flow field distribution. The role of dynamic mesh deformation technology is to automatically update the geometric boundary of the computational grid used in CFD simulation in real time according to the change of dust thickness. Specifically, at each preset dust accumulation time step, At the end of the CFD flow field calculation, the above-mentioned dynamic evolution model of dust accumulation is first used to calculate the dust accumulation thickness increment on each wall unit of the reactor inner wall according to the average particle deposition and scouring conditions within the time step. .

[0053] The calculated dust accumulation thickness increments for each wall element are then applied to the wall boundaries of the CFD model using a suitable dynamic mesh deformation method. Preferred dynamic mesh deformation methods include spring-based smoothing or radial basis function (RBF) interpolation.

[0054] The spring relaxation method treats the edges of the computational grid as virtual springs. The displacement of the wall nodes (caused by the increase in dust thickness) will drive the movement of the internal grid nodes connected to them through the transmission of spring force until the entire grid system reaches a new force balance state, thereby adapting to the changing boundary.

[0055] The radial basis function interpolation method is based on the displacement of known boundary nodes. By constructing an RBF interpolation function, the displacement of internal mesh nodes is calculated. This method can usually maintain the mesh quality well and is especially suitable for cases with large deformations.

[0056] After each mesh deformation operation, it is crucial to check the quality of the updated computational mesh. Mesh quality parameters, such as minimum orthogonality quality, maximum element aspect ratio, and maximum element tilt or twist, must be maintained within acceptable limits to ensure the stability and accuracy of subsequent CFD calculations. If mesh quality deteriorates significantly due to excessive deformation, it may be necessary to trigger a local remeshing or even a global re-meshing.

[0057] The core of this step is to simulate the dynamic coupling of flow field, dust accumulation, and geometry. This is an iterative, time-stepping, cyclical process that tightly integrates CFD flow field calculations, calculations of the dust accumulation dynamic evolution model, and the application of dynamic mesh deformation techniques to capture the complex interactions and temporal evolution of dust accumulation and flow field. This coupled simulation process is executed in a time-stepping loop to obtain continuous, time-varying operating data of the reactor under the influence of the dust accumulation process.

[0058] S3. Using continuous time-varying operating data, define and quantify the time-domain performance indicators of the SCR denitrification reactor within one or more preset operating cycles;

[0059] The implementation of S3 is:

[0060] This step follows the continuous time-varying operating data of the SCR denitrification reactor under the influence of dust accumulation obtained in step S2. Its core purpose is to extract and quantify a set of key indicators that can comprehensively and objectively evaluate the comprehensive performance of the reactor within a preset operating cycle (such as a complete cleaning cycle or a longer observation time) from these massive data that dynamically change over time.

[0061] The continuous time-varying operation data comes from the flow field-dust accumulation-geometry dynamic coupling simulation implemented in step S2. These data are recorded in the form of time series over the entire preset total simulation time. The changes of key physical quantities inside the reactor, such as but not limited to: Instantaneous denitrification efficiency , NOx mass flow rate at reactor inlet and outlet NOx mass flow , reactor total system pressure drop And the instantaneous velocity distribution characteristics of the catalyst inlet section (the instantaneous velocity distribution unevenness index can be calculated These raw data provide a solid foundation for the precise quantification of subsequent time-domain performance indicators.

[0062] The time domain performance index is calculated based on the continuous time-varying operation data obtained in step S2 over the entire simulation duration, and may include at least one of the following representative indicators:

[0063] First, the time-averaged denitrification efficiency is expressed as This indicator is intended to evaluate the average nitrogen oxide conversion capacity of the reactor during the entire simulation period, taking into account the effects of ash accumulation and possible cleaning operations. Its calculation formula is as follows:

[0064] ;

[0065] Where, is the time-averaged denitrification efficiency; is the preset total simulation time; For time; for The instantaneous denitrification efficiency at the moment is calculated as follows:

[0066] ;

[0067] Where, for The mass flow rate of NOx at the reactor inlet at the moment; for The mass flow rate of NOx at the reactor outlet at the moment;

[0068] By measuring the instantaneous denitrification efficiency By integrating and averaging over a period of time, we can obtain a comprehensive index that can reflect the long-term denitrification performance of the reactor. The higher the index is, the better the average working effect of the reactor under dust accumulation conditions.

[0069] Second, the maximum system pressure drop during the cycle is recorded as This indicator is designed to capture the peak value of system resistance caused by gradual blockage of the flow path due to dust accumulation during the entire simulation cycle. The system pressure drop is directly related to the energy consumption of the induced draft fan or forced draft fan. Its maximum value is an important basis for fan selection and evaluation of operating economy. Its calculation formula is as follows:

[0070] ;

[0071] Where, for The total system pressure drop of the reactor at the moment; is the maximum system pressure drop during the cycle; For time; The preset total simulation time.

[0072] By finding the maximum instantaneous system pressure drop over the entire time domain, we can determine the maximum operating resistance of the reactor under the most severe dust accumulation or the most unfavorable flow conditions. When optimizing the design, it is usually desirable to keep this indicator as low as possible to reduce energy consumption.

[0073] Third, the time-averaged catalyst inlet velocity distribution unevenness is expressed as This indicator is designed to assess the uniformity of the average velocity distribution of the flue gas as it enters the catalyst layer during the entire simulation period, due to the potential deterioration of the upstream flow field caused by soot accumulation. The uniformity of the catalyst inlet velocity distribution has a significant impact on the catalyst's utilization efficiency, lifespan, and ammonia slip rate. Its calculation formula is as follows:

[0074] , where is the time-averaged catalyst inlet velocity distribution unevenness; is the preset total simulation time; for The calculation formula for the instantaneous catalyst inlet velocity distribution unevenness index at time is:

[0075] ;

[0076] Where, for The standard deviation of the velocity distribution at the catalyst inlet section at time t; for The average velocity at the catalyst inlet cross section at time t; For time.

[0077] By calculating the instantaneous non-uniformity index over the entire By integrating and averaging the flow rate over a period of time, we can obtain an indicator for evaluating the long-term flow field quality. During design optimization, we hope to keep this indicator as small as possible to ensure full and effective utilization of the catalyst, reduce local wear and blockage, and improve the mixing effect of NH3 and NOx.

[0078] S4. Using a multi-objective optimization algorithm based on a surrogate model, with one or more quantified time-domain performance indicators as optimization objectives and the adjustable design variables of the SCR denitrification reactor as optimization inputs, to perform collaborative optimization, thereby obtaining an optimized SCR denitrification reactor design scheme;

[0079] The implementation of S4 is:

[0080] First, it is necessary to clearly define the adjustable design variables and their ranges for the SCR denitrification reactor. These design variables are the operating objects of the optimization algorithm, and their reasonable selection and parameterization are the key to successful optimization. Adjustable design variables can cover the following aspects:

[0081] First, the structural parameters of the SCR denitrification reactor. These parameters directly determine the macroscopic geometry of the reactor and the characteristics of the internal flow channels. For example, they may include key geometric dimensions of the inlet or outlet flue, such as the aspect ratio of the flue cross section, the diffusion angle of the diffuser, the curvature radius of the turning section, etc.; if the design includes guide plates or flow equalization devices, their number, installation position (such as the precise coordinates in the reactor coordinate system), installation inclination angle, airfoil parameters or curvature radius, and the geometric parameters of the flow equalization grid or porous plate (such as porosity, pore size and distribution, plate thickness, etc.) can all be used as design variables. These structural parameters are usually associated with the CAD geometric model established in step S1 through parametric modeling technology, so that changes in the design variables can be automatically mapped to corresponding changes in the geometric model.

[0082] Second, the dust accumulation management strategy parameters. These parameters involve the specific settings of the cleaning measures taken to alleviate the impact of dust accumulation during the operation of the reactor. For example, if soot blowers are used for dust cleaning, their layout parameters (such as the number of soot blowers, the spatial installation position in the reactor, the directionality or coverage of the sound waves) can be used as optimization variables; further, the soot blowing start and stop logic parameters can be designed to dynamically adjust based on the predicted local dust accumulation rate. For example, the critical dust thickness threshold for triggering the soot blowing operation in different areas, the duration of each soot blowing, the frequency or cycle of soot blowing, and the allocation of different soot blowing frequency weight coefficients for different areas (such as areas with significant differences in dust accumulation rates) can be set.

[0083] Third, preset adjustment logic parameters for adjustable components. If the SCR denitrification reactor design includes components that can be dynamically adjusted based on operating conditions, such as an adjustable catalyst module bracket or a movable guide plate, these preset adjustment logic parameters can also be included in the scope of optimized design variables. For example, when the average dust accumulation thickness in the reactor or the total system pressure drop reaches a certain preset level, the adjustable component (such as the catalyst module bracket) can automatically adjust its tilt angle or translation position to a target setpoint.

[0084] After determining the adjustable design variables and their value ranges, the multi-objective optimization algorithm based on the surrogate model is implemented. This process usually consists of two main stages: the construction of the surrogate model and the iterative optimization stage based on the surrogate model.

[0085] Phase 1: Building the proxy model:

[0086] The goal of this stage is to establish a mathematical model that can quickly predict the time-domain performance indicators (i.e., the indicators quantified in step S3) corresponding to a given design variable combination at a low computational cost. The specific process of building the surrogate model is as follows:

[0087] The first step is to conduct a Design of Experiments (DoE). Within the multidimensional parameter space of predefined adjustable design variables, an efficient DOE method is used to select a set of representative initial sample points. Preferred DOE methods include Latin Hypercube Sampling (LHS), Sobol Sequential Sampling, or uniform design. These methods can achieve good spatial filling and uniformity within the design space, allowing the selected sample points to better represent the characteristics of the entire design space. Each sample point represents a specific combination of adjustable design variables.

[0088] Next, an initial training dataset is generated for each initial sample point selected through the experimental design. Specifically, for each design variable combination corresponding to each sample point (i.e., a specific SCR reactor design), the dynamic coupled flow field, dust accumulation, and geometry simulation (step S2) must be fully executed, followed by the time-domain performance quantification process (step S3). This series of high-precision simulations and calculations yields a set of realistic (i.e., derived through complex numerical simulations) time-domain performance metrics corresponding to that design variable combination. While this process is computationally intensive, it provides high-quality training data for building an accurate and reliable surrogate model.

[0089] After obtaining the initial training data set (i.e., the design variable combinations and their corresponding true performance index values), one or more surrogate models are selected and trained. Preferably, a Kriging model (also known as Gaussian Process Regression) is used as the surrogate model. The Kriging model is an interpolation method that performs a new, un-simulated design point (i.e., a new set of design variable vectors) It can be calculated by the following formula:

[0090] ;

[0091] Where, is The vector of regression functions evaluated at ; is the coefficient estimation vector of the regression function; is the time domain performance indicator value vector observed in the initial training data set; is the regression function matrix evaluated at the sample points of the initial training dataset; It is a correlation matrix that describes the correlation between the response values ​​of each sample point in the initial training data set; It is a vector that describes the correlation between the new design point and each sample point in the initial training data set.

[0092] The above formula provides a predicted value for the unknown point and also provides a measure of uncertainty in that predicted value, namely the prediction variance. This is of great significance for guiding exploration and utilization in the subsequent optimization search process. In addition to the kriging model, other surrogate model options include radial basis function networks, polynomial response surface models, support vector regression, and artificial neural networks. Based on the characteristics of the optimization problem and the data, the most appropriate surrogate model or combination of them can be selected.

[0093] Phase 2:

[0094] Iterative optimization based on proxy models:

[0095] After constructing a preliminary surrogate model, it can be used to perform efficient multi-objective optimization searches. This iterative optimization process aims to continuously improve the accuracy of the surrogate model and gradually approach the true Pareto optimal solution set. The specific steps are as follows:

[0096] First, a multi-objective optimization algorithm is run on the trained surrogate model. Because the evaluation cost of surrogate models is extremely low (typically orders of magnitude faster than real simulation), a large number of function evaluations and iterative optimization searches can be performed at this stage. Common multi-objective optimization algorithms include, but are not limited to, the non-dominated sorting genetic algorithm II (NSGA-II), the multi-objective particle swarm optimization algorithm, and the multi-objective evolutionary strategy. Taking NSGA-II as an example, it simulates the selection, crossover, and mutation operations of biological evolution and combines non-dominated sorting with crowding distance calculation mechanisms to effectively search for and maintain a Pareto approximate solution set in the multi-objective space. This solution set contains a series of non-inferior design solutions that make different trade-offs between different optimization objectives.

[0097] Next, a filling criterion, also known as a point-adding criterion, needs to be designed and implemented to guide the selection of one or a few new design variable combinations as "high-value evaluation points" from the current set of Pareto approximate solutions obtained by the surrogate model, or from the entire design space, so that computational resources can be invested in the realistic, high-cost simulation evaluation steps S2-S3. The filling criterion aims to strike a balance between "exploration" of the surrogate model (i.e., sampling in regions of greater uncertainty to improve the model's global accuracy) and "exploitation" (i.e., sampling in regions of current good prediction performance in the hope of finding even better solutions). Common filling criteria include maximizing expected improvement, which takes into account both the current optimal solution and the uncertainty of the prediction. For multi-objective optimization problems, hypervolume improvement-based criteria can also be employed, which involves selecting points that maximize the hypervolume increment dominated by the current Pareto front. Other criteria, such as improvement probability and confidence lower bounds, can also be used based on the specific problem.

[0098] Then, for the new design variable combination selected based on the filling criteria, the flow field, dust accumulation, and geometry dynamic coupled simulation (step S2) is fully repeated, followed by the time-domain performance index quantification process (step S3) to obtain the actual time-domain performance index value. This step verifies the surrogate model's predictions and is key to obtaining new information to improve the surrogate model.

[0099] Next, the new design variable combinations obtained in the previous step, along with their corresponding real-world performance metric values, are added to the original training dataset to form an expanded dataset. This expanded dataset is then used to retrain or update the surrogate model. For example, for a kriging model, this means reestimating the regression coefficients and hyperparameters of the correlation function and updating the correlation matrix. This model update allows the surrogate model to better fit all known information, including the new data point, thereby improving the accuracy of subsequent predictions.

[0100] Finally, the above steps are repeated until the preset termination conditions are met, such as reaching the total allowed computing resource budget (such as the maximum number of real simulations) or the optimization process converges (such as the Pareto front no longer improves significantly in several consecutive iterations, or the prediction accuracy of the proxy model has reached the required level).

[0101] After the optimization iterations are complete, a final Pareto approximate solution set is obtained, resulting from multiple real-world simulation validations and iterative optimization of the surrogate model. Each solution in this set represents a non-inferior design solution that achieves a specific balance between different time-domain performance indicators. Design engineers can select one or more of the most satisfactory solutions from this Pareto solution set as the final optimized SCR denitrification reactor design based on actual project requirements, cost constraints, and preferences for different performance indicators (for example, whether average denitrification efficiency or maximum pressure drop within a cycle is more important).

[0102] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An SCR denitrification reactor optimization method based on CFD flow field simulation is characterized in that: The following steps are involved: S1. Establishing an initial computational fluid dynamics model of the SCR denitration reactor, and performing a benchmark CFD simulation on the initial computational fluid dynamics model based on preset boundary conditions and a physical model to obtain performance data of the reactor in an initial clean state; S2. Based on the established initial computational fluid dynamics model and the performance data in the initial clean state, a dynamic evolution model of dust accumulation is constructed to describe the change of dust thickness on the inner wall of the reactor over time. The geometry of the initial computational fluid dynamics model is dynamically updated in combination with dynamic mesh deformation technology. A flow field-dust accumulation-geometry dynamic coupling simulation is performed to obtain continuous time-varying operating data of the reactor under the influence of the dust accumulation process. S3. Using the continuous time-varying operating data, define and quantify time-domain performance indicators of the SCR denitration reactor within one or more preset operating cycles; S4. A multi-objective optimization algorithm based on an agent model is used to perform collaborative optimization with one or more of the quantified time-domain performance indicators as optimization targets and the adjustable design variables of the SCR denitrification reactor as optimization inputs, thereby obtaining an optimized SCR denitrification reactor design scheme.

2. The SCR denitrification reactor optimization method based on CFD flow field simulation according to claim 1, characterized in that: In step S1, the step of establishing the initial computational fluid dynamics model includes: A precise three-dimensional geometric model of the reactor is constructed using three-dimensional computer-aided design software, and computational meshing is performed on the three-dimensional geometric model, with local encryption performed in areas where the flow field changes dramatically or the geometry is complex; The preset physical models include a turbulence model, a chemical reaction model and a discrete phase model; The boundary conditions at least include: defining the flue gas velocity or mass flow rate, temperature, pressure and concentration of each gas component at the reactor inlet, defining the pressure at the reactor outlet, and defining the no-slip and temperature conditions of the solid wall of the reactor; The performance data obtained by performing a benchmark CFD simulation on the initial computational fluid dynamics model includes velocity field, temperature field, pressure field, NH3 / NOx molar ratio distribution and NOx conversion rate in the initial clean state obtained through post-processing analysis.

3. The SCR denitrification reactor optimization method based on CFD flow field simulation according to claim 2, characterized in that: In step S2, the dust accumulation dynamic evolution model calculates the wall microelement by the following formula: Thickness of dust accumulation Over time The rate of change of , thus forming the dust accumulation dynamic evolution model: ； Where, for Momentary wall element Thickness of dust accumulation; is the effective density of dust accumulation; for Momentary wall element particle mass deposition flux on the for Momentary wall element Mass scouring or stripping flux of upper dust accumulation; The particle mass deposition flux The calculation is based on the wall surface obtained in the current CFD simulation step in the flow field-dust accumulation-geometry dynamic coupling simulation. Mass concentration of particles within a specific size range in nearby local smoke and particles hit the wall The average normal velocity component of , and combined with the preset effective adhesion probability ; The mass washout or stripping flux The calculation is based on the wall shear stress obtained in the current CFD simulation step in the flow field-dust accumulation-geometry dynamic coupling simulation. Combined with the preset bonding strength of the dust layer .

4. The SCR denitrification reactor optimization method based on CFD flow field simulation according to claim 3, characterized in that: In step S2, the dynamic mesh deformation technology is to calculate the dust thickness increment of each wall unit in each dust accumulation time step based on the dust accumulation dynamic evolution model, and use the spring relaxation method or the radial basis function interpolation method to update the geometric wall boundary of the initial computational fluid dynamics model to form a new geometric boundary, and check and maintain the quality of the computational mesh after each mesh deformation.

5. The SCR denitrification reactor optimization method based on CFD flow field simulation according to claim 4, characterized in that: In step S2, the simulation of the flow field-dust accumulation-geometry dynamic coupling is performed by executing the following time-stepping loop to obtain the continuous time-varying operation data: Performing a CFD simulation under the initial computational fluid dynamics model with the updated geometry at the current moment to calculate and output the flow field, temperature field, component concentration field, particle transport characteristics, and wall parameters used to calculate dust accumulation at that moment; The dust accumulation dynamic evolution model and the wall parameters are used to calculate the dust accumulation time interval. The increment of dust accumulation thickness of each inner wall unit; Using the dynamic mesh deformation technology, the geometric boundary of the initial computational fluid dynamics model is updated according to the calculated dust accumulation thickness increment to form the geometric shape at the next moment; If the simulation time reaches the preset cleaning cycle, the dust accumulation thickness in the specific area will be reduced or removed according to the preset cleaning model; Repeat the above steps until the preset total simulation time is reached .

6. The SCR denitrification reactor optimization method based on CFD flow field simulation according to claim 5, characterized in that: In step S3, the time domain performance index is obtained based on the entire simulation time obtained in step S2. The continuous time-varying operating data is calculated and includes at least one of the following: Time average denitrification efficiency ; Where, is the time-averaged denitrification efficiency; is the preset total simulation time; For time; for The instantaneous denitrification efficiency at the moment is calculated as follows: ； Where, for The mass flow rate of NOx at the reactor inlet at the moment; for The mass flow rate of NOx at the reactor outlet at the moment; Maximum system pressure drop during the cycle ; Where, for The total system pressure drop of the reactor at the moment; is the maximum system pressure drop during the cycle; For time; is the preset total simulation time; Time-averaged catalyst inlet velocity distribution unevenness , where is the time-averaged catalyst inlet velocity distribution unevenness; is the preset total simulation time; for The calculation formula for the instantaneous catalyst inlet velocity distribution unevenness index at time is: ； Where, for The standard deviation of the velocity distribution at the catalyst inlet section at time t; for The average velocity at the catalyst inlet cross section at time t; For time.

7. The SCR denitrification reactor optimization method based on CFD flow field simulation according to claim 6, characterized in that: In step S4, the first stage of adopting the multi-objective optimization algorithm based on the surrogate model is to construct the surrogate model, which includes: Selecting a set of initial sample points in the parameter space of the adjustable design variables using an experimental design method; For each design variable combination represented by the initial sample point, the flow field-dust accumulation-geometry dynamic coupling simulation of step S2 is performed, and the quantization process of step S3 is performed to obtain the time domain performance index value corresponding to the design variable combination to form an initial training data set; Using the initial training data set, a proxy model is selected and trained. The proxy model is a Kriging model, which is used to establish an approximate mathematical relationship between the adjustable design variables and the time domain performance index. The predicted value of the response Calculated by the following formula: ； Where, is The vector of regression functions evaluated at ; is the coefficient estimation vector of the regression function; is the time domain performance indicator value vector observed in the initial training data set; is the regression function matrix evaluated at the sample points of the initial training dataset; It is a correlation matrix that describes the correlation between the response values ​​of each sample point in the initial training data set; It is a vector that describes the correlation between the new design point and each sample point in the initial training data set.

8. The SCR denitration reactor optimization method based on CFD flow field simulation according to claim 7, characterized in that: In step S4, the second stage of the multi-objective optimization algorithm based on the surrogate model is to perform iterative optimization based on the constructed surrogate model, which includes: Running a non-dominated sorting genetic algorithm II on the trained surrogate model to perform a large number of iterative optimizations, thereby quickly obtaining a set of Pareto approximate solutions based on the surrogate model predictions at a very low computational cost; Design and adopt a filling criterion to maximize the expected improvement or a criterion based on super volume improvement, and select one or a few new design variable combinations as the points to be evaluated from the obtained Pareto approximate solution set; For the selected new design variable combination, the flow field-dust accumulation-geometry dynamic coupling simulation of step S2 is performed, and the quantization process of step S3 is performed to obtain its true time domain performance index value; Adding the obtained new design variable combination and its corresponding real time-domain performance index value to the initial training data set, and retraining or updating the proxy model; The above steps are repeated until a preset computing resource budget is reached or the optimization process converges, and finally one or more solutions in the Pareto approximate solution set are output as the optimized SCR denitration reactor design scheme.

9. The SCR denitrification reactor optimization method based on CFD flow field simulation according to claim 8, characterized in that: In step S4, the adjustable design variables of the SCR denitration reactor include at least one of the following: The structural parameters of the SCR denitration reactor are selected from at least one of the inlet flue shape parameters, the outlet flue shape parameters, the number of guide plates, the installation position of the guide plates, the inclination angle of the guide plates, the curvature radius of the guide plates, and the geometric parameters of the flow equalization device; Soot accumulation management strategy parameters are selected from at least one of the following: layout parameters of soot blowers, soot blowing start and stop logic parameters dynamically adjusted based on predicted local soot accumulation rates, and weight coefficients of soot blowing frequencies in different areas; The preset adjustment logic parameters of the adjustable component are selected from the corresponding adjustment inclination angle or translation position of the adjustable bracket of the catalyst module when the average dust accumulation thickness or pressure drop reaches a preset level.

Images