A pulverized coal combustion simulation method based on furnace partition flow field reconstruction

Abstract

The application discloses a pulverized coal combustion simulation method based on a furnace partition flow field reconstruction, comprising the following steps: dividing a furnace into multiple reconstruction partitions; based on computational fluid dynamics (CFD) snapshot data, offline constructing a local reduced basis of each reconstruction partition; according to the physical measurement principle of different sensors, establishing an observation mapping relationship between the full-field variables of the reconstruction partition and the actual measured values of the furnace actual measuring points; checking the partition observable matrix formed by the observation mapping relationship and the local reduced basis, calculating the physical residual of the reconstruction partition after the checking succeeds, and assimilating and updating the coefficients of the local reduced basis by using the physical residual and the observation mapping relationship, so as to obtain the reconstruction state of each reconstruction partition, realize the abnormal prompt and auxiliary diagnosis of the burner, and solve the problems that the traditional reduced model loses local characteristics and the assimilation result does not satisfy the law of conservation of energy, thereby realizing the second-level real-time reconstruction of the three-dimensional flow field of the furnace and the combustion state diagnosis while ensuring the physical high fidelity.

Description

Technical Field

[0001] This invention relates to the field of computer simulation technology, and in particular to a method for simulating pulverized coal combustion based on furnace partition flow field reconstruction. Background Technology

[0002] Pulverized coal industrial boilers and power plant boilers, as large and complex thermal equipment, involve multiphase flow, heat and mass transfer, and intense chemical reactions during the pulverized coal combustion process. To achieve real-time monitoring and optimized control of boiler operation, computational fluid dynamics (CFD)-based computer simulation technology has been widely applied. However, traditional full-size high-fidelity CFD models consume enormous computational resources, making it difficult to meet the real-time requirements of industrial sites. In recent years, methods combining reduced-order models such as intrinsic orthogonal decomposition (POD) with data assimilation have gradually become research hotspots. These methods attempt to utilize offline constructed low-dimensional flow field basis functions, combined with online sparse measurement data, to achieve rapid reconstruction of the three-dimensional flow field inside the furnace.

[0003] However, existing flow field reconstruction methods still have some limitations in practical engineering applications. First, traditional global order reduction models struggle to characterize the high-gradient physical evolution of local regions within the furnace (e.g., the burner near-zone and burnout air zone), and are prone to smoothing out key local features due to insufficient spatial resolution. Second, existing data assimilation and reconstruction methods largely rely on pure mathematical projection or statistical interpolation, lacking the constraints of physical conservation laws. This leads to non-physical phenomena in the reconstructed flow field at local regions or interfaces that do not satisfy the conservation of mass, momentum, and energy. Furthermore, the actual furnace interior has extremely sparse measurement points and suffers from measurement noise and flue gas transport lag. Traditional assimilation algorithms struggle to guarantee the convergence and robustness of the reconstructed state when faced with multi-source, heterogeneous, and highly uncertain observational data, failing to provide highly reliable global state support for burner health diagnosis and operational optimization. Summary of the Invention

[0004] To address the problems of existing flow field reconstruction methods, such as the tendency to smooth out key local features, lack of physical conservation constraints, and difficulty in ensuring the convergence and robustness of the reconstructed state, this invention provides a pulverized coal combustion simulation method based on furnace partition flow field reconstruction.

[0005] This invention is achieved through the following technical solution: A pulverized coal combustion simulation method based on furnace zone flow field reconstruction includes: The furnace is divided into multiple reconstructed partitions, and a local reduced-order basis for each reconstructed partition is constructed offline based on computational fluid dynamics simulation snapshot data. Based on the physical measurement principles of different sensors at actual furnace measuring points, an observational mapping relationship is established between the full-field variables of the reconstructed partition and the measured values ​​of the actual furnace measuring points. The observability of the partition observable matrix formed by the observation mapping relationship and the local reduced order basis is checked. After the check passes, the physical residual of the reconstructed partition is calculated. The reduction coefficient of the local reduced order basis is assimilated and updated using the physical residual of the reconstructed partition and the observation mapping relationship to obtain the optimal reduction coefficient. The state of each reconstructed partition is reconstructed using the optimal reduction coefficient to obtain the reconstructed flow field of each reconstructed partition. Based on the reconstruction flow field of each reconstruction zone, output the furnace reconstruction flow field.

[0006] Preferably, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, dividing the furnace into multiple reconstruction partitions specifically includes: Based on the physical structure of the furnace, the interior of the furnace is divided into multiple initial zones; Obtain physical indices from offline computational fluid dynamics simulation snapshot data, calculate the variance of physical indices in each initial partition, and split the initial partitions whose physical indices variance exceeds a preset variance threshold into multiple sub-partitions. Calculate the flux residual at the interface between adjacent initial partitions, adjacent sub-partitions, or initial partitions and sub-partitions. Merge adjacent initial partitions, adjacent sub-partitions, or initial partitions and sub-partitions with flux residuals lower than a preset flux threshold to form multiple reconstructed partitions.

[0007] Preferably, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, the step of constructing a local reduced-order basis for each reconstructed partition based on computational fluid dynamics simulation snapshot data specifically includes: Extract the state vectors of each reconstructed partition from the offline computational fluid dynamics simulation snapshot data under different working conditions, and construct the local snapshot matrix of each reconstructed partition based on the state vectors; Calculate the average flow field of each reconstructed partition under all operating conditions, and subtract the average flow field from the snapshot matrix to obtain the fluctuation snapshot matrix; Singular value decomposition is performed on the fluctuation snapshot matrix of each reconstructed partition. Local basis functions corresponding to each singular value are extracted as spatial modes, and the state evolution coefficients corresponding to each spatial mode are determined. The top n spatial modes with the highest energy proportion are retained, and the retained spatial modes constitute the local reduced-order basis of the corresponding reconstructed partition.

[0008] Preferably, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, establishing the observation mapping relationship between the full-field variables of the reconstructed partition and the measured values ​​of actual measuring points in the furnace specifically includes: A multi-source data coordinate system for the furnace is constructed, and the coordinates of the CFD simulation mesh and actual measurement points of the furnace are all projected into this multi-source data coordinate system. A flue gas transport influence matrix from each reconstructed zone of the furnace to the furnace outlet section is also established. Observation operators are established according to sensor type, and the observation operators are multiplied with the local reduced basis of the corresponding reconstruction partition to obtain the partition observability matrix; the observability of the partition observability matrix is ​​verified, and when the preset observability conditions are met, the full-field variables of the reconstruction partition are mapped to simulated observation values ​​through the observation operators; For the furnace outlet measuring point, a residence time kernel function is constructed. The residence time kernel function is then discretized in the time domain, integrated over intervals, and normalized to obtain the discrete weight coefficients corresponding to each historical moment. The state vector of the upstream reconstructed partition at each historical moment is multiplied by the corresponding flue gas transport influence matrix, and then weighted and summed with the corresponding discrete weight coefficients. Finally, this summation is added to the simulated observation value of the reconstructed partition corresponding to the furnace outlet measuring point to obtain the simulated observation value of the furnace outlet measuring point at the current moment.

[0009] Preferably, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, calculating the physical residual of the reconstructed partition includes: Construct the physical control equations within each reconstructed partition, substitute the reconstructed flow field corresponding to the current reduction coefficient into the physical control equations, and calculate the internal physical residuals of the reconstructed partitions. For the interface between adjacent reconstruction partitions, the flux continuity deviation of physical quantities on both sides of the interface is calculated to obtain the boundary physical residual of the reconstruction partition. The internal physical residuals and boundary physical residuals of the reconstructed partition together constitute the physical residuals of the reconstructed partition.

[0010] Preferably, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, the order reduction coefficients of the local order reduction basis are assimilated and updated using the physical residuals of the reconstructed partitions and the observation mapping relationship to obtain the reconstructed flow field of each reconstructed partition, specifically including: An assimilation objective function is established that includes a data bias term and a physical constraint term. The data bias term is the data fitting error between the simulated observation value and the measured value corresponding to the actual measurement point, calculated based on the observation mapping relationship. The physical constraint term consists of the physical residuals of the reconstructed partition. Based on the data fitting error and the physical residual of the reconstructed partition, a gating coefficient is calculated for each reconstructed partition. The gating coefficient is used to dynamically adjust the weighting coefficients of the data bias term and the physical constraint term in the assimilation objective function. The weight coefficients are updated using the gating coefficients, and the assimilation objective function is solved iteratively. In each iteration, the reduction coefficients of each reconstructed partition are adjusted with the goal of minimizing the assimilation objective function until the preset iteration termination condition is met, so as to obtain the optimal reduction coefficients of each reconstructed partition. The optimal reduction coefficients are then used to reconstruct the state of each reconstructed partition to obtain the reconstructed flow field of each reconstructed partition.

[0011] Furthermore, in the pulverized coal combustion simulation method based on furnace zone flow field reconstruction, after obtaining the reconstructed flow field of each reconstructed zone, it also includes dual-field coupling reconstruction of the particulate phase and the gas phase: Based on the reconstructed flow field of each reconstructed partition, the three-dimensional gas phase flow field on the CFD full field grid is restored. The Lagrange random orbit model is applied to release the coal powder particle group in the restored CFD full field grid. The mass exchange source term, momentum exchange source term, and energy exchange source term generated by the volatile matter release, coke combustion, and gas-solid drag are calculated. The mass exchange source term, momentum exchange source term, and energy exchange source term are respectively added to the physical control equation of the gas phase to update the physical residuals of each reconstructed partition. The updated physical residuals of each reconstructed partition are then substituted back into the assimilation objective function to perform a second correction of the order reduction coefficients. This process is repeated iteratively until the change in gas phase flow field parameters between two adjacent iterations is less than the preset convergence tolerance, thus obtaining the final furnace reconstructed flow field.

[0012] Preferably, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, after obtaining the furnace reconstructed flow field, the method further includes: Calculate the reliability of the reconstructed flow field for each reconstructed partition; If the confidence level of the reconstructed flow field of a certain reconstructed partition is lower than the preset confidence level threshold, the reconstructed partition is marked as a region to be reviewed, and an abnormal operating condition diagnosis prompt is output.

[0013] Furthermore, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, on the basis of meeting the credibility requirements, the reconstruction feature parameters of each burner outlet region are extracted from the final furnace reconstruction flow field, and compared in real time with the nominal design parameters given in the DCS system to calculate the operating deviation value between the two; it is then determined whether the operating deviation value exceeds the preset safety alarm threshold. If the operating deviation value corresponding to a certain burner exceeds the safety alarm threshold, an abnormal pulverized coal conveying or burner inlet boundary deviation prompt is output.

[0014] Furthermore, in the pulverized coal combustion simulation method based on furnace partition flow field reconstruction, an inversion model is constructed using reconstruction feature parameters. This model uses the burner's inlet boundary conditions as independent variables and minimizes the sum of squared physical residuals of multiple reconstruction partitions as the objective function. The inversion model is solved using the adjoint matrix method or genetic algorithm to invert the actual inlet boundary conditions. If the deviation between the inverted actual inlet boundary conditions and the design parameters provided by the DCS system exceeds a safety threshold, an abnormal inlet boundary condition warning is output for the corresponding burner.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention divides the furnace into multiple reconstruction zones and constructs a local reduced-order basis offline for each zone. This transforms the original ultra-high-dimensional flow field calculation problem, requiring hundreds of thousands of grid nodes, into a low-dimensional optimization problem that only requires solving a small number of reduced-order coefficients online. While reducing the dimensionality of the online assimilation solution, it ensures that the low-dimensional subspace of each zone can tightly represent the unique local flow field structure of that zone. This avoids the problem of insufficient spatial resolution in traditional global single-reduction-order-gene methods, which smooths out key flow field details in high-gradient regions such as the burner near-field and burnout wind zones. By establishing an observational mapping relationship between the overall field variables and the measured values ​​of sparse measuring points according to the physical measurement principles of different sensors, the calculation method of simulated observations matches the physical response mechanism of actual sensors. This avoids the model errors introduced by uniformly simplifying various sensors to grid point interpolation, providing a physically realistic fitting benchmark for assimilation updates. By verifying the partitioned observability matrix, it is ensured that the reduced-order coefficients to be solved under sparse measurement points can be uniquely and stably determined, avoiding underdetermined assimilation problems caused by insufficient observation information. Based on this, the reduced-order coefficients are updated using physical residuals and observation mapping relationships instead of pure mathematical fitting, ensuring that the reconstructed flow field strictly adheres to physical conservation laws while fitting measured data. Therefore, this invention's partitioned reduction provides a low-dimensional subspace for observability verification and assimilation updates, physical mapping provides a measured data bridge for assimilation, and physical residuals provide conservation constraints for assimilation. Ultimately, it achieves simultaneous improvements in computational efficiency, reconstruction accuracy, and physical fidelity, and can directly output a highly reliable three-dimensional reconstructed flow field of the furnace that meets the real-time requirements of industrial sites.

[0016] Furthermore, this invention divides the initial partitions according to the physical structure of the furnace, and then performs adaptive splitting based on the variance of physical indicators in the offline simulation snapshot. When the variance of physical indicators in a certain initial partition exceeds a preset threshold, it indicates that the physical gradient inside the region is large and the flow field structure is complex. A single reduced-order basis is insufficient to characterize its entire flow field morphology. At this time, by splitting into multiple sub-regions, the flow field characteristics in each sub-region tend to be uniform, thereby ensuring that the local reduced-order basis extracted from each sub-region can achieve high-precision reconstruction with a compact number of modes, avoiding the problem of insufficient approximation accuracy of the reduced-order basis due to drastic changes in the flow field within the partition. At the same time, by calculating the flux residuals at the interfaces between adjacent initial partitions, adjacent sub-regions, or initial partitions and sub-regions, and merging regions with flux residuals lower than a preset flux threshold, adjacent regions with highly consistent flow field characteristics and no need for independent characterization are merged into the same reconstruction partition. This avoids unnecessary partition refinement and the resulting increase in online computational burden, while ensuring the physical continuity and consistency on both sides of the partition boundary.

[0017] Furthermore, when constructing local reduced-order basis offline, this invention calculates and subtracts the average flow field of each partition under all operating conditions to obtain a fluctuation snapshot matrix that only reflects the flow field fluctuations and operating condition changes. Then, it extracts the local basis functions corresponding to each order of singular values ​​as spatial modes through singular value decomposition and truncates and retains the main spatial modes according to the energy contribution rate. This allows the low-dimensional subspace of each reconstructed partition to compactly represent the unique flow field structure of that partition, thereby minimizing the degree of freedom of online solution while ensuring reconstruction accuracy.

[0018] Furthermore, this invention constructs a unified coordinate system for multi-source furnace data, projecting the CFD simulation grid and the actual measurement point coordinates onto the same coordinate system, thus eliminating spatial positional deviations between different data sources. Based on this, observation operators are constructed according to the physical measurement principles of sensors, ensuring that the calculation method of simulated observations matches the physical response mechanism of actual sensors, thereby improving the physical realism of the data fitting terms. Simultaneously, addressing the issue of flue gas transport lag at the outlet measurement point, a dynamic time-delay observation mapping relationship is constructed, integrating the observation operators established according to sensor physical principles with the time-dwelling kernel function. Through the flue gas transport influence matrix and the dwell time kernel function, the spatiotemporal accumulation and delay impact of the historical state of the upstream main combustion zone on the current observation value of the downstream outlet measurement point is quantitatively characterized, clarifying the causal time-delay relationship between upstream and downstream variables and ensuring the convergence of the assimilation algorithm under dynamic operating conditions.

[0019] Furthermore, this invention calculates the internal physical residuals by substituting the current reconstructed flow field into the physical control equations within each partition, and calculates the flux continuity deviation at the interface between adjacent partitions as the boundary physical residuals, which together constitute a complete physical residual term. This ensures that the reconstructed flow field strictly follows the conservation laws of mass, momentum, energy, and components while fitting sparse measurement point data during the assimilation and update process, eliminating numerical jumps and non-physical patterns at the partition boundaries and improving the physical fidelity of the reconstructed flow field.

[0020] Furthermore, this invention constructs an assimilation objective function that integrates data bias terms and physical constraint terms, and dynamically adjusts the weight coefficients of data bias terms and physical constraint terms in the assimilation objective function according to the relative magnitude of data fitting error and physical residual by using gating coefficients. This achieves noise-resistant and robust assimilation updates and effectively avoids assimilation divergence caused by measurement noise.

[0021] Furthermore, after reconstructing the gas phase flow field, this invention calculates the mass, momentum, and energy exchange source terms generated by the volatile matter release of pulverized coal particles, coke combustion, and gas-solid drag using a Lagrange random orbit model. These source terms are then fed back into the gas phase physical control equations for secondary correction of the order reduction coefficients. Through the alternating iteration of the gas phase field and the particle phase field, the dual-field synchronous reconstruction of the multiphase strongly coupled combustion process is achieved. This enables the reconstructed flow field to accurately reflect the momentum, energy, and mass transfer between pulverized coal particles and the gas phase, further enhancing the ability of the reconstruction results to characterize the real combustion process.

[0022] Furthermore, this invention quantifies and evaluates the reliability of the reconstructed flow field in each reconstructed partition, and marks the corresponding partition as a region to be reviewed and outputs an abnormal operating condition diagnosis prompt when the reliability is lower than a preset threshold. This avoids outputting an unreliable reconstructed flow field when the current operating condition deviates significantly from the offline snapshot operating condition space, thus ensuring the reliability and engineering usability of the reconstruction results.

[0023] Furthermore, this invention utilizes the reconstructed flow field to extract the reconstructed characteristic parameters of each burner outlet region, compares them in real time with the DCS nominal design parameters, and calculates the operating deviation value. When the operating deviation value exceeds the limit, it outputs a coal powder conveying abnormality or inlet boundary deviation prompt, providing a specific and operable quantitative basis for the health diagnosis of boiler burners.

[0024] Furthermore, this invention constructs an inversion model with the burner inlet boundary conditions as the independent variable and minimizing the sum of squared physical residuals of multiple zones as the objective function. By inverting the actual inlet boundary conditions through the adjoint matrix method or genetic algorithm, it can quantitatively identify inlet velocity deviation, pulverized coal concentration deviation, or particle size offset anomaly. The inversion deviation is then used to compensate and correct adjustment commands such as coal feed rate and primary air volume, realizing a complete technical chain from flow field reconstruction to anomaly diagnosis and closed-loop optimization control. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating the overall process of a pulverized coal combustion simulation method based on furnace partition flow field reconstruction, as described in one embodiment of the present invention. Detailed Implementation

[0027] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0028] It should be noted that the process equipment or apparatus not specifically mentioned in the following embodiments are all conventional equipment or apparatus in the art.

[0029] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Furthermore, unless otherwise stated, the numbering of each method step is merely a convenient tool for identifying each method step, and not intended to limit the order of the method steps or define the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0030] Example 1 Reference Figure 1 This is the first embodiment of the present invention, which provides a pulverized coal combustion simulation method based on furnace partition flow field reconstruction, including: S1. Divide the furnace into multiple reconfiguration zones and construct the local reduced-order basis for each reconfiguration zone offline based on computational fluid dynamics (CFD) snapshot data.

[0031] It should be noted that, in order to address the problem that traditional global order reduction methods tend to lose local flow field details when dealing with local high gradients and intense chemical reactions within the furnace (e.g., in the burner region), this invention employs a partitioned order reduction strategy to reduce the computational dimensionality while locking in and preserving the characteristic flow field structure of each region.

[0032] Furthermore, based on the internal geometry of the power plant boiler furnace, the intensity of the combustion reaction, and the fluid dynamics characteristics, the entire three-dimensional space within the furnace is divided into several independent reconfiguration zones. The process of dividing these reconfiguration zones is as follows: First, multiple initial zones are defined based on the furnace's physical structure (e.g., burner elevation, flame deflection angle, outlet flue location, etc.). Then, physical parameters such as temperature and velocity are acquired from offline computational fluid dynamics simulation snapshot data, and the variance of these physical parameters within each initial zone is calculated. If the variance of physical parameters within a given initial zone exceeds a preset variance threshold, it indicates that the physical gradient within that initial zone is too large and the flow field is complex. In this case, the initial zone is split into multiple smaller sub-zones using a grid bisection method or clustering algorithm. Finally, the flux residuals (e.g., the average difference in mass flow rate or heat flux) at the interfaces between adjacent initial zones, adjacent sub-zones, and between initial zones and sub-zones are calculated. If the flux residuals between adjacent initial zones, adjacent sub-zones, and between initial zones and sub-zones are below a preset flux threshold, it indicates that the flow characteristics of these two zones are highly consistent, and these two adjacent zones are merged.

[0033] Specifically, through the adaptive iteration of splitting and merging described above, multiple independent reconstructed partitions are ultimately formed. The reconstructed partition refers to the computational region ultimately used for subsequent local reduced-order basis construction and online flow field reconstruction after initial partitioning, necessary sub-region splitting, and merging of adjacent partitions. Therefore, the reconstructed partition can originate from the initial partition without splitting, from the split sub-regions, or from a composite partition formed by merging after meeting the flux continuity requirement. In this embodiment, the furnace is divided into the following three typical partitions: Burner zone (main combustion zone): In this zone, pulverized coal mixes violently with primary and secondary air and ignites and burns. The flow rate, temperature and component concentration gradients are extremely high, and it belongs to a strongly nonlinear reaction zone.

[0034] Burnout air zone (OFA zone): This zone introduces a large proportion of burnout air to suppress the generation of nitrogen oxides. There is a strong jet mixing and secondary combustion process, and the flow field shear force is large.

[0035] Upper part of the furnace and outlet area (dilute phase burnout zone): Combustion is basically completed in this area, the flow tends to be gentle, and the main process is convective heat transfer of high-temperature flue gas, with a relatively uniform flow field distribution.

[0036] It should be noted that this partitioning method can transform complex global nonlinear flow into several local flow problems with relatively simple physical characteristics, making it easier to extract the flow field characteristics of each partition in a targeted manner.

[0037] Furthermore, in order to enable the constructed local reduced-order basis to cover various operating conditions of the boiler in actual operation, this invention uses computational fluid dynamics (CFD) software to perform numerical simulation of the full-size furnace.

[0038] Specifically, firstly, key operating parameters such as boiler load (e.g., 100%, 75%, 50% load), coal quality characteristics (e.g., volatile matter, calorific value), and air supply ratio (e.g., primary air rate, burnout air offset angle) are changed, and dozens to hundreds of steady-state or transient CFD simulations are run. After the simulation converges, the physical quantity data of the internal spatial grid nodes of the furnace under each operating condition are extracted, including three-dimensional velocity vectors, temperature, oxygen concentration, carbon dioxide concentration, and pulverized coal particle concentration. Finally, according to the preset reconstruction partition boundaries, the physical quantity data of the global CFD grid nodes are segmented and classified, thus obtaining the state vector of each reconstruction partition. For each reconstruction partition, the physical quantity data of all operating conditions are concatenated column-wise to form a local snapshot matrix specific to that reconstruction partition.

[0039] It is important to emphasize that the above-mentioned process of partitioning, splitting sub-regions, and merging adjacent partitions is completed through iterative optimization during the offline data preprocessing stage. Once the optimal partition boundaries are determined, the spatial boundaries of each partition will remain fixed during the subsequent online assimilation process to ensure the matching of the local reduced-order basis space dimensions.

[0040] Furthermore, for the local snapshot matrix of each reconstructed partition, the intrinsic orthogonal decomposition method is used for spatial dimensionality reduction and feature extraction. The specific process is as follows: The average flow field of each reconstructed partition under all operating conditions is calculated, and this average flow field is subtracted from each original local snapshot matrix to obtain a fluctuation snapshot matrix that only reflects flow field fluctuations and operating condition changes. Then, singular value decomposition (SVD) is performed on this fluctuation snapshot matrix (e.g., using Singular Value Decomposition (SVD) or Orthogonal Eigenvalue Decomposition (POD); Dynamic Mode Decomposition (DMD) can also be used when extracting unsteady evolution features is required). Local basis functions corresponding to each order of singular values ​​are extracted as spatial modes, and the state evolution coefficients corresponding to each spatial mode are determined. It should be explained that this decomposition process is physically equivalent to decomposing complex flow field fluctuations into a series of spatially orthogonal local basis functions (i.e., spatial modes) and state evolution coefficients used to characterize the weights of the corresponding spatial modes as a function of time or operating state. Finally, all obtained singular values ​​are arranged in descending order, and the cumulative energy contribution rate of the spatial modes corresponding to the first few singular values ​​(i.e., the proportion of the sum of the first n singular values ​​to the total sum of all singular values) is calculated. When the cumulative energy contribution rate reaches a preset threshold (e.g., 99% or 99.5%), the top n spatial modes with the highest energy proportion are retained, while the remaining high-frequency noise modes are discarded. This set of retained spatial modes constitutes the local reduced-order basis of the reconstructed partition.

[0041] Furthermore, after the above processing, the local flow field of any reconstructed partition under any specific operating condition can be approximated by a linear combination of the average flow field and the local reduced-order basis of that reconstructed partition. Its physical reconstruction relation is expressed as follows:

[0042] in, For the first The order reduction coefficients corresponding to each spatial mode.

[0043] It should be noted that in the above reconstruction relationship, the number of grid nodes (degrees of freedom) in each reconstruction partition is usually as high as hundreds of thousands or even millions, while the number of spatial modes n retained is usually extremely small (generally less than 15). Therefore, through the order reduction process in this step, the ultra-high-dimensional flow field calculation problem that originally required solving hundreds of thousands of grid nodes can be successfully transformed into a low-dimensional optimization problem that only requires solving a dozen or so unknown reduction coefficients online, laying the application foundation for realizing second-level real-time reconstruction of furnace flow field.

[0044] It should be further clarified that this invention does not directly solve the three-dimensional multivariable full field of each grid node in the furnace from extremely sparse measurement points. Instead, it solves for a finite number of reduction coefficients within a low-dimensional local reduction subspace formed by offline CFD snapshots. For the k-th reconstruction partition, the observation operator of the reconstruction partition is first multiplied by the local reduction basis to form the partition observability matrix, and the column rank, condition number, or minimum singular value is checked on the partition observability matrix. When the partition observability matrix satisfies the preset observability conditions, i.e., the column rank is not less than the number of modes in the retained space, the condition number is not greater than the preset condition number threshold, or the minimum singular value is not less than the preset singular value threshold, the sparse measurement point data, combined with physical residuals, interface flux continuity constraints, and historical state priors, are used to uniquely or stably determine the reduction coefficients of the reconstruction partition. If a reconstructed partition does not meet the preset observability condition, observability is restored by reducing the number of retained spatial modes, merging adjacent partitions, increasing the weight of physical constraints, introducing prior reduction coefficients from the previous time step, or adding acoustic / optical survey lines. Partitions that still do not meet the condition are marked as regions to be verified. Therefore, the solution object of this method is the low-dimensional reduction coefficients that satisfy the observability check, rather than unconstrained inversion of the complete high-dimensional field from a very small number of measurement points, thus avoiding underdetermined solutions.

[0045] S2. Based on the physical measurement principles of different sensors at the actual measuring points in the furnace, establish the observation mapping relationship between the full-field variables of the reconstructed partition and the measured values ​​of the actual measuring points in the furnace.

[0046] It should be noted that in the actual operation of power plant boilers, the high temperature, high corrosiveness, and strong reducing atmosphere inside the furnace result in an extremely sparse arrangement of physical sensors (such as thermocouples, acoustic thermometers, and flue gas analyzers), and the measurement data are often accompanied by significant background noise and physical lag. Based on this, this invention constructs a high-fidelity observation operator that maps the full-field variables of each reconstructed partition in the high-dimensional furnace to an extremely sparse actual measurement point space, providing an accurate comparison benchmark for data assimilation and updates.

[0047] Furthermore, in order to eliminate spatial positional deviations between different data sources (such as boiler design drawings, CFD grids, and on-site DCS measurement point coordinates), this invention constructs a unified coordinate system for multi-source furnace data.

[0048] Specifically, a three-dimensional Cartesian coordinate system is established with the furnace zero point (e.g., the center of the furnace bottom or a specific corner reference point) as the origin. The spatial envelope positions of the boiler's geometric boundaries, the center coordinates of each burner nozzle, the cross-sections of secondary air and burnout air nozzles, the heating surfaces (e.g., water-cooled walls, screen-type superheaters), and the physical coordinates of various actual measuring points installed on-site (e.g., wall thermocouples, acoustic temperature measurement lines, outlet flue gas sampling guns) are all projected into this unified coordinate system. Then, using a spatial proximity interpolation algorithm, the specific reconstruction partition of each actual measuring point in the CFD mesh is determined, and its corresponding mesh node index is locked, achieving seamless alignment between the physical space and the simulation space.

[0049] Furthermore, for the transient physical field inside the furnace, sensor readings are affected not only by the current local state but also by measurement noise. Therefore, this invention establishes a mapping relationship between the global variable vector of each reconstructed partition and the measured values ​​of the actual measuring points in the furnace by constructing a spatial observation operator matrix. It should be noted that the spatial observation operator matrix is ​​used to map the global variable vector to the simulated observation values ​​corresponding to the actual measuring points; the partition observable matrix is ​​obtained by multiplying the spatial observation operator matrix with the local reduced-order basis of the corresponding reconstructed partition, and is used to determine whether sparse measuring points are sufficient to stably solve for the reduced-order coefficients. The two have different functions but are related.

[0050] Specifically, this invention does not simplify all sensors to a single grid-point trilinear interpolation, but instead establishes observation operators according to the physical measurement principles of the sensors. For point-type measuring points that directly measure local gas components or velocities within the furnace, the corresponding rows in the spatial observation operator matrix can be set with non-zero values ​​at several grid nodes closest to the actual measuring point's physical coordinates, and the simulated observation value of the local gas variable is obtained based on the interpolation coefficients. For thermocouples arranged on water-cooled walls or heated surfaces, the observed object is the pipe wall or heated surface wall temperature. The observation operator maps the near-wall gas temperature, wall heat flux, pipe wall thermal resistance, and cooling medium boundary conditions together to a simulated observation value of the wall temperature, instead of directly replacing the thermocouple reading with the trilinear interpolation of the grid gas temperature within the furnace. For acoustic temperature sensors, the observation operator represents the integral of the sound propagation time along the sound wave propagation path, that is, integrating the reciprocal of the sound velocity related to the local gas temperature and composition along the sound ray path, and forming a simulated observation value from this propagation time or the calibrated path average temperature. For TDLAS (Dielectric Laser Absorption Spectroscopy) sensors, the observation operator represents the absorbance integral along the beam path, which is determined by local temperature, pressure, analyte concentration, spectral line intensity, and line shape function. Finally, observation noise vectors are introduced into various observation operators, and covariance matrices are set according to sensor accuracy, installation location, and operating condition disturbances to reasonably adjust the weights of different measurement data during assimilation.

[0051] Furthermore, due to the limited flue gas velocity inside the furnace, the flue gas components (e.g., nitrogen oxides, carbon monoxide) and heat generated from the main combustion zone (upstream) require several seconds of physical transport before reaching the actual measuring point at the furnace outlet (downstream). Directly assimilating the current outlet measurement value with the current main combustion zone state would lead to serious non-causal bias. Therefore, this invention introduces a residence time kernel function and a flue gas transport influence matrix to construct a dynamic time-delay observation mapping relationship.

[0052] Specifically, the simulated observation value at the furnace outlet measuring point is defined as the sum of two parts: the first part is the local simulated observation value obtained by mapping the local state of the reconstructed partition corresponding to the furnace outlet measuring point through the observation operator; the second part is the cumulative value of the state vector of each upstream reconstructed partition (e.g., main combustion zone, burnout air zone) at historical moments, after time-domain convolution integration using the flue gas transport influence matrix and the residence time kernel function. This flue gas transport influence matrix is ​​constructed based on offline CFD streamline tracing technology, and its matrix elements characterize the proportion of mass and energy contribution of the flue gas flow rate of a specific upstream reconstructed partition to different spatial locations of the furnace outlet cross-section. Meanwhile, due to the presence of turbulent mixing and backflow within the furnace, the time for flue gas molecules to be transported from the upstream of the furnace to the furnace outlet is not a single fixed value, but a probability distribution. Based on this, this embodiment preferably models the residence time kernel function as a log-normal distribution function, whose scale parameter is determined by the average geometric distance from the upstream reconstructed partition to the furnace outlet and the average flue gas velocity along this path, and whose shape parameter is used to characterize the intensity of turbulent diffusion and mixing of the flue gas during transport. It should be explained that the residence time kernel function quantitatively gives the attenuation and delay law of the influence of the historical state of the upstream reconstructed partition on the current furnace outlet measuring point.

[0053] It should be noted that the flue gas transport influence matrix is ​​constructed as follows: By releasing a large number of massless particle streams at each upstream reconstructed partition grid, the probability distribution of the particle swarm reaching different measuring points at the downstream furnace outlet section from upstream and the weight ratio of the mass and energy it carries are statistically analyzed, thereby obtaining the matrix elements of the flue gas transport influence matrix. This matrix quantitatively characterizes the contribution of the state change of a specific upstream reconstructed partition to the response of measuring points at different spatial locations at the furnace outlet.

[0054] In one possible implementation of this embodiment, the expression for the residence time kernel function represented by the log-normal distribution function is as follows:

[0055] Among them, scale parameter The average physical transport time from upstream of the furnace to the furnace outlet was obtained based on flowline tracing. Sure( The average geometric distance from the upstream of the furnace to the furnace outlet. (where is the average flue gas velocity from upstream of the furnace to the furnace outlet), and their relationship is: Shape parameters The empirical mapping of turbulent kinetic energy integrals of the corresponding path in the CFD snapshot is obtained. The larger the turbulent kinetic energy, the larger the shape parameter, indicating that the axial mixing of flue gas is more intense during the transport process.

[0056] Furthermore, in order to facilitate real-time online calculations in the computer control system, the aforementioned continuous integral form of the lag observation mapping relationship is discretized in the time domain.

[0057] Specifically, a sampling time interval consistent with the online monitoring system is set, and the maximum physical transport delay is divided into a finite number of discrete time delay steps. Then, at each discrete time node, the residence time kernel function is integrated over an interval and normalized to obtain the discrete weight coefficients corresponding to each historical moment. Finally, the state vector of the upstream reconstruction partition at each historical sampling moment is multiplied by the corresponding flue gas transport influence matrix, and then weighted and summed with the corresponding discrete weight coefficients. Finally, this sum is added to the local simulated observation value of the reconstruction partition corresponding to the furnace outlet measuring point to output the simulated observation value of the furnace outlet measuring point at the current moment.

[0058] It should be noted that, through the above-mentioned dual spatial and temporal mapping, this invention not only establishes the geometric relationship between sparse measurement points and high-dimensional full-field variables in space, but also straightens out the causal time delay relationship between upstream and downstream variables in the temporal dimension through the residence time kernel function. This ensures that subsequent assimilation can accurately and reasonably feed back and compensate the measurement deviation at the furnace outlet to the corresponding upstream reconstruction partition when facing dynamic operating conditions.

[0059] S3. Perform observability verification on the partition observable matrix formed by the observation mapping relationship and the local reduced basis, calculate the physical residual of the reconstructed partition, and use the physical residual and the observation mapping relationship to assimilate and update the coefficients of the local reduced basis to obtain the reconstructed flow field of each reconstructed partition.

[0060] Furthermore, physical control equations are constructed within each reconstructed partition, including mass conservation equations, momentum conservation equations, energy conservation equations, and component transport equations.

[0061] Specifically, the reconstructed flow field of the reconstructed partition corresponding to the current reduction coefficient is substituted into the physical governing equation, and the deviation on both sides of the equation is calculated to obtain the internal physical residual of the reconstructed partition. Simultaneously, for the interface between adjacent reconstructed partitions, the flux continuity deviation of physical quantities (including pressure, velocity, temperature, and component concentration) on both sides of the interface is calculated to obtain the boundary physical residual. It is important to emphasize that the internal physical residual and the boundary physical residual of the reconstructed partition together constitute the physical residual of the reconstructed partition.

[0062] It should be noted that the calculation method for the boundary physical residuals is as follows: The difference in fluid flux (e.g., mass flux and heat flux calculated based on density and velocity) between two adjacent reconstructed partitions at the common interface is calculated, and the area integral is performed at the interface to serve as a constraint to ensure the continuity of mass and energy conservation across the reconstructed partitions.

[0063] Furthermore, an assimilation objective function is constructed that integrates real-time measurement data with physical conservation laws. This assimilation objective function is composed of a weighted sum of data bias terms and physical constraint terms.

[0064] Specifically, the data deviation term is obtained by using the observation mapping relationship established in S2 to calculate the data fitting error (e.g., mean square error) between the simulated observation value and the measured value of the actual measurement point under the current order reduction coefficient.

[0065] Specifically, the physical constraint term consists of the physical residuals of the reconstructed partitions (including the internal physical residuals and boundary physical residuals of the reconstructed partitions) calculated above.

[0066] Specifically, by setting their respective weighting coefficients in the data deviation term and the physical constraint term, the relative importance between the data fitting accuracy and the physical conservation constraints is balanced.

[0067] In one possible implementation of this embodiment, the assimilation objective function can be expressed as:

[0068] in, Let be the order reduction coefficients to be determined. These are the measured values ​​from actual sparse measurement points. For the observation operator, To reconstruct the physical residuals of the partition. It is the weighting coefficient of the data deviation term. These are the weighting coefficients for the physical constraint terms.

[0069] Furthermore, the gating coefficient for each reconstructed partition is calculated based on the data fitting error and the physical residuals of the reconstructed partitions. The data fitting error is the error amount corresponding to the data deviation term, and the physical residuals of the reconstructed partitions already include internal physical residuals and boundary physical residuals; therefore, the interface throughput difference is no longer considered as a repeating term independent of the physical residuals.

[0070] Specifically, the gating coefficient That is, the weighting coefficients and The dynamic adjustment factor. In this embodiment, the gating coefficient is based on the data fitting error of the q-th reconstructed partition. With physical residual The proportional relationship is calculated, and its expression can be represented as:

[0071] in, To adjust the sensitivity constant, To avoid small positive quantities with a denominator of zero.

[0072] Then, the weight coefficients are updated using this gating coefficient, for example, by letting... , ,in and The initial weighting coefficients are used. In this way, in each iteration, when the data fitting error is normal, the assimilation process considers both the data and the physical model; when the data fitting error surges abnormally (e.g., the sensor is subjected to sudden high-frequency noise interference or even fails), causing the data bias term to be much larger than the physical residual of the reconstructed partition, the gating coefficients tend to 0, at which point the system dynamically increases... The proportion of the forced reconstruction of the flow field follows the physical conservation law, filtering out observation noise.

[0073] It should be noted that this invention achieves noise-resistant robust assimilation through this gating coefficient.

[0074] Furthermore, the assimilation objective function is solved iteratively using an optimization algorithm (such as gradient descent or quasi-Newton method).

[0075] Specifically, in each iteration, the reduction coefficients of each reconstructed partition are dynamically adjusted with the goal of minimizing the assimilation objective function. When the iteration meets the preset termination condition (the value of the assimilation objective function is less than a set threshold, or the preset maximum number of iterations is reached), the iteration stops, and the optimal reduction coefficients of each reconstructed partition are output.

[0076] Furthermore, the optimal reduction coefficients are used to reconstruct the state of each reconstruction partition.

[0077] Specifically, the optimal reduction coefficient of each reconstruction partition is linearly combined with the corresponding local reduction basis (i.e., the optimal reduction coefficient is multiplied and accumulated with the local reduction basis vector), and the combined result is superimposed on the average flow field of the reconstruction partition to obtain the final reconstruction flow field (including temperature field, velocity field and component concentration field, etc.) of each reconstruction partition.

[0078] S4. Output the furnace reconstruction flow field based on the reconstruction flow field of each reconstruction partition.

[0079] It should be noted that, since the pulverized coal combustion process is a typical high-concentration, multiphase, strongly coupled flow, the reconstructed flow field in S3 alone cannot accurately characterize the heterogeneous reaction processes such as the movement, heating, volatile matter release, and coke combustion of pulverized coal particles within the furnace. Therefore, after obtaining the preliminary gas-phase reconstructed flow field of each reconstructed zone, a dual-field coupled reconstruction of the particulate and gas phases must be performed. In this reconstruction process, the present invention employs an alternating iterative approach.

[0080] Specifically, firstly, the three-dimensional gas-phase flow field (including velocity and temperature fields) on the full-size grid is reconstructed using the optimal reduction coefficients of each reconstructed partition. Then, a Lagrange random orbit model is applied to release the coal powder particle swarm in the reconstructed full-field high-dimensional grid space. The mass exchange, momentum exchange, and energy exchange source terms generated by volatile matter release, coke combustion, and gas-solid drag within each grid cell are calculated. Next, these newly calculated mass exchange, momentum exchange, and energy exchange source terms are added to the source terms of the corresponding gas-phase physics governing equations (i.e., mass, momentum, energy, and component transport equations) for each reconstructed partition, updating the physical residuals of the reconstructed partitions. The updated physical residuals of the reconstructed partitions are then substituted back into the assimilation objective function of S3 for a secondary correction of the reduction coefficients. This process is repeated iteratively until the change in gas phase flow field parameters (e.g., temperature change amplitude, velocity change amplitude) between two adjacent iterations is less than the preset convergence tolerance. At this point, it is determined that the dual-field coupling has reached an equilibrium state, and the final furnace reconstructed flow field is output.

[0081] Furthermore, after obtaining the final furnace reconstructed flow field, a reliability assessment of the furnace zone reconstructed flow field is conducted.

[0082] Specifically, this confidence level is quantified based on the comprehensive residual after the assimilation iteration convergence, and the calculation formula can be expressed as: .in, Let be the confidence level of the reconstructed flow field in the q-th reconstructed partition. This is the weighted sum of the physical residuals and data fitting error of the reconstructed partition, where α is the scaling factor. When... If the confidence level is lower than the preset confidence threshold (e.g., 0.8), it indicates that the current local reduced-order basis cannot effectively characterize the extreme abnormal conditions of the reconstructed partition. The system marks the reconstructed partition as a region to be reviewed and outputs abnormal condition diagnosis prompts to avoid outputting false reconstructed flow fields.

[0083] Furthermore, while ensuring that the credibility requirements are met, the reconstructed flow field characteristics are used to provide abnormal indications and auxiliary diagnosis of the boiler burner's operating status.

[0084] Specifically, from the final reconstructed flow field of the furnace, reconstruction characteristic parameters of each burner outlet region are extracted. These parameters include at least: reconstructed pulverized coal concentration, reconstructed primary air velocity, and reconstructed combustion temperature. The extracted reconstruction characteristic parameters are compared in real time with the nominal design parameters given in the distributed control system (DCS), and the operating deviation value between the two is calculated. It is then determined whether this operating deviation value exceeds a preset safety alarm threshold. If the operating deviation value corresponding to a certain burner exceeds the safety alarm threshold, an abnormal pulverized coal delivery or burner inlet boundary deviation warning is output, rather than directly judging a single reconstructed pulverized coal concentration deviation as hardware damage. For example, if the reconstructed flow field shows that the pulverized coal concentration at the outlet of a certain burner corner is lower than 80% of the design value (i.e., deviation exceeds the limit), the system outputs a warning of insufficient pulverized coal delivery or inlet boundary deviation for that corner; subsequently, combined with the corresponding measured results of coal feed rate, pulverized coal pipe pressure difference, primary air velocity, and pulverized coal concentration, it further distinguishes possible fault types such as pulverized coal pipe blockage, primary air volume deviation, or wall coking, and outputs maintenance suggestions for review.

[0085] Furthermore, utilizing the reconstructed characteristic parameters, an inversion model is constructed with inlet boundary conditions (e.g., inlet velocity, tangential velocity, pulverized coal concentration) as independent variables and minimizing the sum of squares of physical residuals (e.g., momentum conservation residuals, mass conservation residuals) across multiple reconstructed zones as the objective function. This inverse optimization problem is solved using the adjoint matrix method or a genetic algorithm to invert the actual inlet boundary conditions. If the deviation between the inverted actual inlet boundary parameters and the DCS design parameters exceeds a safety threshold, an abnormality warning is issued for the corresponding burner's inlet velocity deviation, pulverized coal concentration deviation, or particle size offset. Combined with measured results of coal feed rate, pulverized coal pipe pressure differential, primary air velocity, and pulverized coal concentration, further confirmation is made regarding fault types such as pulverized coal pipe blockage, primary air adjustment deviation, or wall-mounted coking.

[0086] Furthermore, to achieve feedback control between online simulation and actual furnace operation, the system can use the reconstructed characteristic parameters obtained above as the initial flow field for the next round of CFD calculations, thereby shortening the iterative convergence time of the flow field simulation. Alternatively, these parameters can be extracted and introduced into the online monitoring model as boundary correction terms or relaxation source terms. Simultaneously, the inverted inlet boundary condition deviations are compensated to achieve health diagnosis and high-fidelity prediction of the boiler combustion system.

[0087] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for pulverized coal combustion simulation based on furnace zone flow field reconstruction, characterized in that, include: The furnace is divided into multiple reconstructed partitions, and a local reduced-order basis for each reconstructed partition is constructed offline based on computational fluid dynamics simulation snapshot data. Based on the physical measurement principles of different sensors at actual furnace measuring points, an observational mapping relationship is established between the full-field variables of the reconstructed partition and the measured values ​​of the actual furnace measuring points. The observability of the partition observable matrix formed by the observation mapping relationship and the local reduced order basis is checked. After the check passes, the physical residual of the reconstructed partition is calculated. The reduction coefficient of the local reduced order basis is assimilated and updated using the physical residual of the reconstructed partition and the observation mapping relationship to obtain the optimal reduction coefficient. The state of each reconstructed partition is reconstructed using the optimal reduction coefficient to obtain the reconstructed flow field of each reconstructed partition. Based on the reconstruction flow field of each reconstruction zone, output the furnace reconstruction flow field.

2. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 1, characterized in that, The division of the furnace into multiple reconfiguration zones specifically includes: Based on the physical structure of the furnace, the interior of the furnace is divided into multiple initial zones; Obtain physical indices from offline computational fluid dynamics simulation snapshot data, calculate the variance of physical indices in each initial partition, and split the initial partitions whose physical indices variance exceeds a preset variance threshold into multiple sub-partitions. Calculate the flux residual at the interface between adjacent initial partitions, adjacent sub-partitions, or initial partitions and sub-partitions. Merge adjacent initial partitions, adjacent sub-partitions, or initial partitions and sub-partitions with flux residuals lower than a preset flux threshold to form multiple reconstructed partitions.

3. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 1, characterized in that, The construction of a local reduced-order basis for each reconstructed partition based on computational fluid dynamics simulation snapshot data specifically includes: Extract the state vectors of each reconstructed partition from the offline computational fluid dynamics simulation snapshot data under different working conditions, and construct the local snapshot matrix of each reconstructed partition based on the state vectors; Calculate the average flow field of each reconstructed partition under all operating conditions, and subtract the average flow field from the snapshot matrix to obtain the fluctuation snapshot matrix; Singular value decomposition is performed on the fluctuation snapshot matrix of each reconstructed partition. Local basis functions corresponding to each singular value are extracted as spatial modes, and the state evolution coefficients corresponding to each spatial mode are determined. The top n spatial modes with the highest energy proportion are retained, and the retained spatial modes constitute the local reduced-order basis of the corresponding reconstructed partition.

4. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 1, characterized in that, The establishment of the observation mapping relationship between the global variables of the reconstructed partition and the measured values ​​of the actual measuring points in the furnace specifically includes: A multi-source data coordinate system for the furnace is constructed, and the coordinates of the CFD simulation mesh and actual measurement points of the furnace are all projected into this multi-source data coordinate system. A flue gas transport influence matrix from each reconstructed zone of the furnace to the furnace outlet section is also established. Observation operators are established according to sensor type, and the observation operators are multiplied with the local reduced basis of the corresponding reconstruction partition to obtain the partition observability matrix; the observability of the partition observability matrix is ​​verified, and when the preset observability conditions are met, the full-field variables of the reconstruction partition are mapped to simulated observation values ​​through the observation operators; For the furnace outlet measuring point, a residence time kernel function is constructed. The residence time kernel function is then discretized in the time domain, integrated over intervals, and normalized to obtain the discrete weight coefficients corresponding to each historical moment. The state vector of the upstream reconstructed partition at each historical moment is multiplied by the corresponding flue gas transport influence matrix, and then weighted and summed with the corresponding discrete weight coefficients. Finally, this summation is added to the simulated observation value of the reconstructed partition corresponding to the furnace outlet measuring point to obtain the simulated observation value of the furnace outlet measuring point at the current moment.

5. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 1, characterized in that, Calculating the physical residuals of the reconstructed partitions includes: Construct the physical control equations within each reconstructed partition, substitute the reconstructed flow field corresponding to the current reduction coefficient into the physical control equations, and calculate the internal physical residuals of the reconstructed partitions. For the interface between adjacent reconstruction partitions, the flux continuity deviation of physical quantities on both sides of the interface is calculated to obtain the boundary physical residual of the reconstruction partition. The internal physical residuals and boundary physical residuals of the reconstructed partition together constitute the physical residuals of the reconstructed partition.

6. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 1, characterized in that, The reduction coefficients of the local reduction basis are assimilated and updated using the physical residuals of the reconstructed partitions and the observation mapping relationship to obtain the reconstructed flow field of each reconstructed partition, specifically including: An assimilation objective function is established that includes a data bias term and a physical constraint term. The data bias term is the data fitting error between the simulated observation value and the measured value corresponding to the actual measurement point, calculated based on the observation mapping relationship. The physical constraint term consists of the physical residuals of the reconstructed partition. Based on the data fitting error and the physical residual of the reconstructed partition, a gating coefficient is calculated for each reconstructed partition. The gating coefficient is used to dynamically adjust the weighting coefficients of the data bias term and the physical constraint term in the assimilation objective function. The weight coefficients are updated using the gating coefficients, and the assimilation objective function is solved iteratively. In each iteration, the reduction coefficients of each reconstructed partition are adjusted with the goal of minimizing the assimilation objective function until the preset iteration termination condition is met, so as to obtain the optimal reduction coefficients of each reconstructed partition. The optimal reduction coefficients are then used to reconstruct the state of each reconstructed partition to obtain the reconstructed flow field of each reconstructed partition.

7. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 6, characterized in that, After obtaining the reconstructed flow field of each reconstructed partition, the process also includes dual-field coupling reconstruction of the particulate phase and the gas phase: Based on the reconstructed flow field of each reconstructed partition, the three-dimensional gas phase flow field on the CFD full field grid is restored. The Lagrange random orbit model is applied to release the coal powder particle group in the restored CFD full field grid. The mass exchange source term, momentum exchange source term, and energy exchange source term generated by the volatile matter release, coke combustion, and gas-solid drag are calculated. The mass exchange source term, momentum exchange source term, and energy exchange source term are respectively added to the physical control equation of the gas phase to update the physical residuals of each reconstructed partition. The updated physical residuals of each reconstructed partition are then substituted back into the assimilation objective function to perform a second correction of the order reduction coefficients. This process is repeated iteratively until the change in gas phase flow field parameters between two adjacent iterations is less than the preset convergence tolerance, thus obtaining the final furnace reconstructed flow field.

8. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 1, characterized in that, After obtaining the reconstructed flow field in the furnace, the following is also included: Calculate the reliability of the reconstructed flow field for each reconstructed partition; If the confidence level of the reconstructed flow field of a certain reconstructed partition is lower than the preset confidence level threshold, the reconstructed partition is marked as a region to be reviewed, and an abnormal operating condition diagnosis prompt is output.

9. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 8, characterized in that, Based on the premise that the credibility requirement is met, the reconstruction characteristic parameters of each burner outlet area are extracted from the final furnace reconstruction flow field and compared in real time with the nominal design parameters given in the DCS system to calculate the operating deviation value between the two; it is then determined whether the operating deviation value exceeds the preset safety alarm threshold. If the operating deviation value corresponding to a certain burner exceeds the safety alarm threshold, an abnormal coal powder conveying or burner inlet boundary deviation prompt is output.

10. The pulverized coal combustion simulation method based on furnace zone flow field reconstruction according to claim 9, characterized in that, Using the reconstructed feature parameters, an inversion model is constructed with the burner inlet boundary conditions as independent variables and minimizing the sum of squared physical residuals of multiple reconstructed partitions as the objective function. The inversion model is solved by the adjoint matrix method or genetic algorithm to invert the actual inlet boundary conditions. If the deviation between the actual inlet boundary conditions obtained by inversion and the design parameters provided by the DCS system exceeds the safety threshold, an abnormal inlet boundary condition prompt is output for the corresponding burner.

Images