Temperature field prediction model training method, temperature field reconstruction method

Abstract

This application belongs to the fields of data processing and deep learning, specifically involving a temperature field prediction model training method and a temperature field reconstruction method, aiming to solve the problem of insufficient accuracy in boiler furnace temperature measurement. The temperature field prediction model training method includes: acquiring the geometric model of the target boiler and dividing the geometric model into grids, wherein each grid corresponds to grid data used for positioning; acquiring the temperature field data of the target boiler based on a fluid numerical simulation model; using the target dataset constructed from the grid data and temperature field data as training samples for the unit model, and iteratively training the node temperature update model and edge feature update model included in the unit model; synchronizing the model parameters of the trained unit model to multiple layers to generate a temperature field prediction model composed of multiple layers. This application can achieve high-precision reconstruction of the internal temperature field of a target boiler using a small amount of measured temperature data.

Description

Technical Field

[0001] This application belongs to the fields of data processing and deep learning, and specifically relates to a method for training a temperature field prediction model and a method for reconstructing a temperature field. Background Technology

[0002] As a crucial piece of equipment in the industrial sector, the operating efficiency and safety of boilers directly impact system energy efficiency, safety, and operational stability. During boiler operation, furnace temperature is a vital parameter characterizing combustion status, fuel utilization efficiency, and safe operation. It can be used to analyze core boiler characteristics such as coking, high-temperature corrosion, nitrogen oxide (NOx) formation, and heat transfer efficiency, thereby optimizing boiler energy efficiency, environmental performance, and safety. Therefore, comprehensive, real-time, and accurate monitoring of boiler furnace temperature is of paramount importance.

[0003] Currently, commonly used methods for monitoring boiler furnace temperature mainly include thermocouple temperature measurement, infrared temperature measurement, and acoustic temperature measurement, or emerging methods such as furnace simulation and temperature field prediction through deep learning.

[0004] However, traditional boiler furnace temperature monitoring methods lack measurement accuracy, are not comprehensive, and cannot perform real-time 3D monitoring. Existing numerical calculation and deep learning methods have low prediction accuracy and require a large amount of measured data as a basis. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, namely the insufficient accuracy of boiler furnace temperature measurement, this application, in its first aspect, proposes a method for training a temperature field prediction model, comprising:

[0006] Obtain the geometric model of the target boiler and mesh the geometric model. The mesh of the geometric model corresponds to the mesh data used for positioning.

[0007] Based on the fluid numerical simulation model of the target boiler, the temperature field data of the target boiler are obtained;

[0008] The target dataset constructed from grid data and temperature field data is used as the training sample for the unit model. The node temperature update model and edge feature update model included in the unit model are trained iteratively.

[0009] The model parameters of the trained unit model are synchronized to multiple layers to generate a temperature field prediction model composed of multiple layers, wherein the structure of the layers is the same as that of the unit model.

[0010] As a preferred implementation, the node temperature update model and edge feature update model included in the unit model are trained iteratively, including:

[0011] Obtain the initial node model and initial edge model of the mesh, where,

[0012] The input data for the initial node model includes the initial node temperature and edge features. The output data of the initial node model is the updated node temperature, and the edge features are either the initial edge features or the updated edge features.

[0013] The input data of the initial edge model includes the initial edge features and the node features corresponding to the nodes connected by the edge. The output data is the updated edge features, where the node features include the three-dimensional coordinates of the grid corresponding to the node, the measured temperature, the initial temperature, the furnace pressure, and the edge features.

[0014] The initial node model is iterated over at the first time step, and the updated node temperature at the second time step is used as the initial node temperature at the second time step to obtain the node temperature update model.

[0015] The updated edge features of the initial edge model at the first time step are used as the initial edge features at the second time step. The initial edge model is iterated to obtain the edge feature update model.

[0016] Obtain the node-edge critical matrix of the mesh, and couple the node-edge critical matrix, the node temperature update model, and the edge feature update model to generate a cell model. The node-edge critical matrix is ​​used to characterize the correspondence between the nodes and edges of the mesh.

[0017] As a preferred implementation, when the edge feature update model converges, the target edge feature output by the edge feature update model is obtained and used as the edge feature in the input data of the node temperature update model.

[0018] As a preferred implementation method, based on the fluid numerical simulation model of the target boiler, the temperature field data of the target boiler is obtained, including:

[0019] Based on the fluid numerical simulation model of the target boiler, simulation data of combustion in multiple boiler furnaces were obtained. The simulation data includes boiler furnace geometric parameters, mesh data, and internal temperature field data of the boiler furnace. The simulation data is used to simulate the temperature field distribution under different furnace types and different operating conditions.

[0020] Data augmentation is performed on the temperature field data inside the boiler furnace to obtain enhanced temperature field data;

[0021] The enhanced temperature field data is stitched together with other data from the simulation data to generate the temperature field data of the target boiler.

[0022] As a preferred implementation, based on the fluid numerical simulation model of the target boiler, simulation data of combustion in multiple boiler furnaces are obtained, including:

[0023] Based on fluid numerical modeling and simulation, construct fluid, combustion and heat transfer models in any boiler furnace;

[0024] The fluid, combustion, and heat transfer models are solved to obtain simulation datasets, which can be used as simulation data for combustion in any boiler furnace.

[0025] As a preferred embodiment, data augmentation is performed on the temperature field data inside the boiler furnace to obtain enhanced temperature field data, including:

[0026] The intermediate iteration results of the simulation data in any iteration process are obtained as enhanced temperature field data.

[0027] As a preferred embodiment, data augmentation is performed on the temperature field data inside the boiler furnace to obtain enhanced temperature field data, which further includes:

[0028] Obtain the measured temperature at random node locations;

[0029] The temperature at the random node locations in the intermediate iteration results is replaced with the measured temperature to obtain enhanced temperature field data.

[0030] A second aspect of this application proposes a method for reconstructing a temperature field, characterized by comprising:

[0031] Based on the grid data and local measured temperatures of the target boiler, a measured dataset is constructed;

[0032] The measured dataset is input into the temperature field prediction model to obtain the temperature field prediction result of the target boiler. The temperature field prediction model is trained by the above-mentioned temperature field prediction model training method.

[0033] Based on the measured temperature of the target at the preset location and the predicted temperature field, the temperature field of the target boiler is reconstructed.

[0034] As a preferred implementation, based on the measured temperature of the target location and the predicted temperature field, the temperature field of the target boiler is reconstructed, including:

[0035] Extract the local predicted temperature at a preset location from the temperature field prediction results;

[0036] If the difference between the local predicted temperature and the target measured temperature is less than the deviation threshold, the temperature field prediction result is used as the temperature field reconstruction result of the target boiler.

[0037] If the difference between the local predicted temperature and the target measured temperature is greater than or equal to the deviation threshold, the temperature field prediction result is re-input into the temperature field prediction model for temperature field prediction until the difference between the local predicted temperature extracted from the temperature field prediction result and the target measured temperature is less than the deviation threshold. The output result of the temperature field prediction model is then used as the temperature field reconstruction result of the target boiler.

[0038] As a preferred implementation, the method further includes: acquiring measured datasets from multiple historical moments to form a historical dataset;

[0039] The prediction dataset is obtained based on historical datasets and time series prediction models;

[0040] Input the predicted dataset into the temperature field prediction model to obtain the predicted temperature field reconstruction results.

[0041] The beneficial effects of this application are:

[0042] (1) First, the geometric model of the target boiler is obtained and the geometric model is meshed. The mesh of the geometric model corresponds to the mesh data used for positioning. The three-dimensional shape of the target boiler is displayed by geometric modeling of the target boiler. The mesh is used to cut the set model, which is convenient for data processing in smaller units. Further, based on the fluid numerical simulation model of the target boiler, the temperature field data of the target boiler is obtained. The simulation data of the target boiler is obtained as temperature field data through the simulation model, which can make up for the incomplete detection of the measured temperature data. The target dataset constructed by the mesh data and temperature field data is used as the training sample of the unit model. The node temperature update model and the edge feature update model included in the unit model are trained iteratively. The unit model is trained by iteratively updating the node temperature and edge features, which enables the unit model to simulate the iterative simulation process of the temperature field and learn the simulation process of the temperature field of the target boiler. The model parameters of the trained unit model are synchronized to multiple layers to generate a temperature field prediction model composed of multiple layers. The distributed training method reduces the amount of data processing for model training and improves the efficiency of model training. Attached Figure Description

[0043] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0044] Figure 1 This is a flowchart of a temperature field prediction model training method provided in the first embodiment of this application;

[0045] Figure 2 This is a flowchart of the steps of a unit model training process provided in the first embodiment of this application;

[0046] Figure 3 This is an example diagram of a unit model provided in the first embodiment of this application;

[0047] Figure 4 This is a flowchart of a temperature field reconstruction method provided in the second embodiment of this application;

[0048] Figure 5 This is an example geometric model diagram of a target boiler provided in one embodiment of this application;

[0049] Figure 6 This is a system block diagram of the temperature field prediction model training system provided in the third embodiment of this application;

[0050] Figure 7 This is a system block diagram of the temperature field reconstruction system provided in the fourth embodiment of this application;

[0051] Figure 8 This is a schematic diagram of the boiler furnace temperature field reconstruction system provided in the fifth embodiment of this application. Detailed Implementation

[0052] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0053] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0054] Currently, common boiler furnace temperature monitoring technologies mainly include the following:

[0055] 1. Thermocouple temperature measurement technology:

[0056] A thermocouple is a traditional temperature sensor, usually installed in a specific location inside a boiler furnace, to calculate the temperature by measuring the electromotive force generated by the temperature difference.

[0057] Advantages: The measurement principle is simple, the cost is low, and it is suitable for medium and low temperature regions.

[0058] shortcoming:

[0059] (1) Limited high temperature resistance: Thermocouples are easily damaged in high temperature environments and have a short service life.

[0060] (2) Spatial limitations: Thermocouples can only measure the temperature at a fixed point and cannot cover the temperature distribution inside the furnace.

[0061] (3) Slow response speed: When the temperature changes rapidly, the thermocouple may not be able to reflect the actual situation in time.

[0062] 2. Infrared thermometry technology:

[0063] Infrared thermometry is based on the principle that the infrared energy emitted by an object is related to its temperature, and the temperature is calculated by detecting the intensity of the radiation.

[0064] Advantages: Non-contact temperature measurement, suitable for high-temperature environments, and capable of monitoring a wide range of areas.

[0065] shortcoming:

[0066] (1) Susceptible to flue gas interference: Infrared sensors are easily affected by flue gas, dust and water vapor when the boiler is running, which leads to increased measurement error.

[0067] (2) Complicated calibration: Infrared sensors require regular calibration and have high requirements for monitoring angle and position.

[0068] 3. Fiber optic temperature measurement technology:

[0069] Fiber optic temperature measurement utilizes the optical properties of optical fibers to estimate temperature distribution by measuring the attenuation or scattering of transmitted light.

[0070] Advantages: High temperature resistance, electromagnetic interference resistance, suitable for distributed temperature monitoring in complex environments.

[0071] shortcoming:

[0072] (1) High cost: The manufacturing and maintenance costs of fiber optic temperature measurement equipment are high, making it unsuitable for large-scale promotion.

[0073] (2) Reliability issues: Optical fibers are easily damaged in high temperature, high pressure and corrosive gas environments.

[0074] 4. Sound temperature measurement technology:

[0075] Sound temperature measurement is an emerging technology that uses the principle that the speed of sound in high-temperature gas changes with temperature to estimate the furnace temperature by measuring the propagation time of the sound waves.

[0076] Advantages: Enables non-contact temperature measurement, suitable for high temperature and harsh environments, and can provide distributed temperature data.

[0077] shortcoming:

[0078] (1) Greatly affected by background noise: Combustion noise, mechanical vibration and steam noise in the boiler may interfere with the sound wave signal and affect the measurement accuracy.

[0079] (2) Complex signal processing: The propagation path of sound waves in the furnace is complex and may be affected by airflow disturbances, requiring complex algorithms to restore the signal and calculate the temperature.

[0080] (3) High equipment calibration requirements: Acoustic temperature measurement equipment needs to be calibrated regularly to ensure measurement accuracy.

[0081] 5. Furnace simulation:

[0082] Furnace simulation is a numerical calculation method for fluids. It discretizes the Navier-Stokes equations based on the geometric discrete grid of the furnace and uses finite element, finite volume, or finite difference algorithms to calculate core field data such as temperature, pressure, and flow velocity at or within the grid nodes, thereby achieving the goal of reflecting combustion inside the furnace.

[0083] shortcoming:

[0084] (1) Extremely high computational complexity: For conventional furnace types, the number of grids is between 3 million and 10 million, and the computation time is on the order of hours, which cannot be applied to real-time operation on site;

[0085] (2) Cannot be adjusted in real time: The fluid numerical calculation model is designed with a large number of empirical parameters, which cannot be adjusted in real time according to real-time data;

[0086] 6. Temperature field prediction based on deep learning:

[0087] Predicting physical fields based on deep learning is an emerging method for predicting and calculating temperature fields. By accumulating massive amounts of data and establishing the relationship between boundary conditions and temperature fields based on deep learning models, the temperature field can be predicted in real time, achieving the goal of real-time prediction.

[0088] shortcoming:

[0089] (1) Difficult to correct in real time: Currently, temperature field reconstruction algorithms based on deep learning methods are all static supervised learning methods. They establish the relationship between boundary parameters, auxiliary feature variables and target temperature field, but do not effectively combine local measured temperature field data to improve the overall model prediction effect.

[0090] (2) No spatial correlation of temperature field data points: The current training process is to predict and evaluate single points of temperature field data, which is difficult to reflect the spatial relationship of temperature field measurement points and is prone to distortion of temperature field spatial characteristics.

[0091] This application provides a method for training a temperature field prediction model and a method for reconstructing a temperature field. It can achieve high-precision reconstruction of the internal temperature field of a target boiler with a small amount of measured temperature data, while taking into account both spatial characteristics and the accuracy of the temperature field.

[0092] To more clearly explain the training method of the temperature field prediction model in this application, the following will be combined with... Figure 1 The steps in the embodiments of this application are described in detail.

[0093] The temperature field prediction model training method of the first embodiment of this application includes steps S110-S140, each of which is described in detail below:

[0094] Step S110: Obtain the geometric model of the target boiler and perform mesh generation on the geometric model, wherein the mesh of the geometric model corresponds to the mesh data used for positioning.

[0095] Optionally, the geometric model is three-dimensional in order to fully reflect the three-dimensional distribution of the temperature field inside the target boiler furnace.

[0096] As one possible implementation, an ensemble model is established using fluid simulation software, such as using ANSYS Fluent to create a geometric model of the target boiler.

[0097] It should be noted that the boiler structure may differ for different furnace types, and an adaptive geometric model needs to be established for the target boiler to be reconstructed in temperature field.

[0098] Furthermore, the geometric model can be meshed using meshing tools that support Adaptive Mesh Refinement (AMR), such as the HyperMesh tool.

[0099] It should be noted that AMR is a technique for dynamically adjusting the computational grid density, which aims to improve accuracy in critical areas by locally refining the grid, while maintaining coarse granularity in non-critical areas to save computational resources.

[0100] The geometric model is divided into multiple meshes using a meshing tool, where each mesh is a three-dimensional geometric block.

[0101] It should be noted that when using a mesh generation tool, the mesh type and density need to be determined based on the actual operating conditions of the target boiler in order to determine the optimal combination of computational load and accuracy, and to balance computational complexity and accuracy.

[0102] Optionally, structured or unstructured meshes can be used during the mesh generation process. For example, structured meshes use a hexahedral structure, while unstructured meshes use a tetrahedral structure.

[0103] Optionally, the grid data can be grid 3D coordinates that enable the positioning of nodes within the grid.

[0104] Step S120: Based on the fluid numerical simulation model of the target boiler, obtain the temperature field data of the target boiler.

[0105] Optionally, based on the fluid numerical simulation model of the target boiler, simulation data of combustion in multiple boiler furnaces can be obtained: Based on fluid numerical modeling and simulation, a fluid, combustion and heat transfer model in any boiler furnace can be constructed; the fluid, combustion and heat transfer model can be solved to obtain a simulation calculation dataset, which can be used as the simulation data of combustion in any boiler furnace. The simulation data includes boiler furnace geometric parameters, mesh data, and internal temperature field data of the boiler furnace. The simulation data is used to simulate the temperature field distribution under different furnace types and different operating conditions.

[0106] In the embodiments of this application, the physical models include, but are not limited to, thermodynamic property models, turbulence models (such as k-ε, k-omega and related extended models), combustion chemical reaction kinetic models (based on fuel chemical composition), and radiation heat transfer models (P-1 or DOM models, etc.), combustion models, etc.

[0107] In this embodiment, boundary conditions are set and processed, such as primary air volume or velocity, secondary air volume or velocity, fuel feed rate or velocity, and heat exchange boundary on the steam-water side. The boundary conditions need to be expanded based on measured data.

[0108] As an example, the measured data is the detection data of the temperature, pressure and flow sensors acquired by the distributed control system.

[0109] Furthermore, data augmentation is performed on the simulation data to expand the simulation dataset.

[0110] Optionally, acquire simulation data of combustion in multiple boiler furnaces; perform data augmentation on the temperature field data inside the boiler furnace to obtain augmented temperature field data; and stitch the augmented temperature field data with other data in the simulation data to generate the temperature field data of the target boiler.

[0111] As an example, the intermediate iteration results of the simulation data in any iteration process are obtained as enhanced temperature field data.

[0112] As another example, the measured temperature at random node locations is obtained; the temperature at the random node locations in the intermediate iteration results is replaced with the measured temperature to obtain enhanced temperature field data.

[0113] As another example, the initial temperature field is randomly initialized. Random initialization strategies include, but are not limited to, uniform initialization with random values ​​and random initialization with a given random distribution, which improves the model's generalization and iteration capabilities under random initialization.

[0114] As another example, the initialization strategy for intermediate iteration results of the initial temperature field is to use multiple sets of intermediate calculation results from simulation iterations to initialize the temperature field, approximating the combustion initialization effect at different time points.

[0115] As another example, random sampling of the convergence result space and iterative random noise are used to simulate the measured temperature field, avoiding the input loss problem caused by the lack of measured temperature field data; at the same time, large-sample space sampling increases the model's generalization ability to different temperature sensor installation locations.

[0116] Step S130: The target dataset constructed from grid data and temperature field data is used as the training sample for the unit model. The node temperature update model and edge feature update model included in the unit model are trained iteratively.

[0117] Optional, combined Figure 2 The flowchart below shows the steps for training the unit model:

[0118] Step S131: Obtain the initial node model and initial edge model of the mesh.

[0119] The input data for the initial node model includes the initial node temperature and edge features. The output data of the initial node model is the updated node temperature, and the edge features are either the initial edge features or the updated edge features. The input data for the initial edge model includes the initial edge features and the node features corresponding to the nodes connected by the edge. The output data is the updated edge features.

[0120] Optionally, the input data for the initial node model is a 2nd-order tensor, with the tensor dimension being the number of nodes N. v and the number of node features N e That is, the tensor is an N v ×N e The matrix, the node features are the initial node temperature corresponding to the grid node and various temperature-related parameters during the combustion process in the boiler furnace, including at least the initial node temperature and edge features. In the embodiments of this application, it includes, but is not limited to, the three-dimensional coordinates of the node, whether the temperature is measured, the measured temperature value, the initial node temperature, the furnace pressure, the furnace composition, whether it is a boundary, the boundary type, the boundary quantity, the edge features and other variables.

[0121] The initial node temperature can be a randomly initialized temperature, a uniform temperature value determined based on experience, or a predicted temperature value from the previous moment; the edge features are edge latent space features; the boundary types are Diriclet boundary type (represented by 0 in the node features) and Neumann boundary type (represented by 1 in the node features); the boundary quantity is the quantified data of the operating parameters, such as fuel velocity, heat transfer density, fuel composition, etc.; the edge features are the combination of the edge features of each edge associated with the node.

[0122] In this embodiment, the initial node model has a graph neural network (GNN) structure. The matrix composed of the node features of each node in the grid is input into the initial node model as input data. The initial node model is processed by batch normalization and fully connected layers to output updated node temperature. The updated node temperature is a prediction of the initial node temperature. The loss function is trained by the difference between the updated node temperature and the true label. The loss function can be mean squared error loss, mean absolute error loss, etc.

[0123] Optionally, the input to the edge model is edge features and node combination features. Edge features include, but are not limited to, edge vectors, edge lengths, adjacent node feature vectors, and the ratio of edge length to boiler feature length. Node combination features are combinations of node features corresponding to the nodes connected by the edge features.

[0124] In this embodiment of the application, the model structure of the edge model is also a graph neural network (GNN). The edge features and node combination features are combined as input data into the edge model. The model is processed by a fully connected layer and the updated edge features are output. The updated edge features are a prediction of the initial edge features. The loss function is trained by the difference between the updated edge features and the true label. The loss function can be the mean squared error loss, the mean absolute error loss, etc.

[0125] It should be noted that the real label is the simulation data obtained from the furnace combustion simulation of the target boiler.

[0126] Step S132: Use the updated node temperature of the initial node model at the first time step as the initial node temperature at the second time step, and iterate the initial node model to obtain the node temperature update model.

[0127] Optionally, the updated node temperature of each node in the grid at the first time step is replaced with the corresponding initial node temperature, which is then used as the initial node temperature at the second time step. This updated node temperature is combined with other node features to form a new matrix, which is then used as the input data for the initial node model. The loss function of the initial node model is then trained, and the initial node model is iterated to obtain the node temperature update model.

[0128] Step S133: Use the updated edge features of the initial edge model at the first time step as the initial edge features at the second time step, and iterate the initial edge model to obtain the edge feature update model.

[0129] Optionally, the updated edge features of each node in the grid at the first time step are used to replace the corresponding initial edge features as the initial edge features at the second time step. These updated edge features are then combined with the corresponding node combination features to form new input data. The loss function of the initial edge model is then trained again, and the initial edge model is iterated to obtain the edge feature update model.

[0130] Step S134: Obtain the node-edge critical matrix of the mesh, and couple the node-edge critical matrix, the node temperature update model, and the edge feature update model to generate a cell model. The node-edge critical matrix is ​​used to characterize the correspondence between nodes and edges.

[0131] Optionally, obtain the correspondence between each node and each edge within the grid. For example, if edge k is not connected to node j, then the corresponding relationship coefficient M is obtained. kj =0, otherwise M kj These are non-zero learnable relation coefficients, and the node-edge critical matrix is ​​composed of the relation coefficients within the grid.

[0132] Furthermore, by coupling the node-edge critical matrix, the node temperature update model, and the edge feature update model, a unit model is generated, which can be expressed by the following formula:

[0133]

[0134] in, Let x represent the i-th unit model. i This represents the input data for the i-th unit model. Represents the initial node model. To represent a composite function, This indicates the number of iterations for the initial node model. Iteration The node temperature update model is then updated. Denotes the initial edge model, x' i This represents the input data for the initial edge model. This represents the number of iterations for the initial edge model. Iteration The edge feature update model after the next step.

[0135] As an example, such as Figure 3 The example diagram of the unit model shown illustrates that the initial edge model can be iteratively updated first. Once the edge feature update model converges, the target edge features output by the edge feature update model are obtained and used as edge features in the input data of the node temperature update model. In other words, the output of the updated edge feature update model is used as a parameter in the node features of the node temperature update model, and the node temperature update model is iteratively updated.

[0136] Step S140: The model parameters of the trained unit model are synchronized to multiple layers to generate a temperature field prediction model composed of multiple layers, wherein the structure of the layers is the same as that of the unit model.

[0137] Optionally, the results parameters of the trained unit model can be copied to multiple layers with the same structure as the unit model to generate a temperature field prediction model composed of multiple layers, which can further iterate the unit model.

[0138] The temperature field prediction model is designed as a multi-layer iterative structure with consistent structure in each layer. By replicating the unit model to different layers, the iterative capability of the unit model can be improved, thus increasing the model convergence efficiency.

[0139] As one possible implementation, the model consisting of multiple layers can be trained until convergence to obtain a temperature field prediction model.

[0140] This application first obtains the geometric model of the target boiler and then meshes the geometric model. Each mesh corresponds to a grid for positioning. Geometric modeling of the target boiler displays its three-dimensional shape. Mesh partitioning cuts the model into smaller units for data processing. Further, based on the fluid numerical simulation model of the target boiler, temperature field data is obtained. Using simulation data from the simulation model as temperature field data compensates for the incompleteness of measured temperature data. The target dataset constructed from the mesh data and temperature field data is used as training samples for the unit model. The node temperature update model and edge feature update model included in the unit model are iteratively trained. Iterative updates of node temperature and edge features enable the unit model to simulate the iterative simulation process of the temperature field and learn the simulation process of the target boiler's temperature field. The model parameters of the trained unit model are synchronized to multiple layers to generate a temperature field prediction model composed of multiple layers. Distributed training reduces the amount of data processing required for model training and improves training efficiency.

[0141] Please see Figure 4 The second embodiment of this application provides a temperature field reconstruction method, based on the above-mentioned temperature field prediction model training method, the specific steps of which include:

[0142] Step S410: Based on the grid data and local measured temperature of the target boiler, construct a measured dataset.

[0143] Optionally, the local measured temperature can be obtained at the sampling point selected for the target boiler. For example, the local measured temperature can be obtained through a sensor, and the local measured temperature can be stitched together with the grid data to generate a measured dataset.

[0144] As a possible implementation, the measured dataset may also include node 3D coordinates, whether the temperature is measured, the measured temperature value, the initial node temperature, furnace pressure, furnace composition, whether it is a boundary, boundary type, boundary quantity, edge characteristics, and other variables.

[0145] Step S420: Input the measured dataset into the temperature field prediction model to obtain the temperature field prediction result of the target boiler.

[0146] Optionally, the measured data set consisting of the actual temperatures of the target boiler can be input into the temperature field prediction model to obtain a more accurate temperature of the target boiler at the current moment.

[0147] Understandably, the temperature field prediction results are temperature predictions made on the entire geometric model of the target boiler, and can describe the temperature distribution inside the target boiler.

[0148] Step S430: Based on the measured temperature of the target at the preset location and the predicted temperature field, the temperature field of the target boiler is reconstructed.

[0149] Optionally, the target measured temperature at a preset location can be obtained through a sensor, and the target measured temperature can be used to verify and correct the temperature field prediction results to obtain the temperature field reconstruction results.

[0150] In this embodiment, the local predicted temperature at a preset location is extracted from the temperature field prediction result; if the difference between the local predicted temperature and the target measured temperature is less than the deviation threshold, the temperature field prediction result is used as the temperature field reconstruction result of the target boiler; if the difference between the local predicted temperature and the target measured temperature is greater than or equal to the deviation threshold, the temperature field prediction result is re-input into the temperature field prediction model for temperature field prediction until the difference between the local predicted temperature extracted from the temperature field prediction result and the target measured temperature is less than the deviation threshold, and the output result of the temperature field prediction model is used as the temperature field reconstruction result of the target boiler.

[0151] Comparing the local predicted temperature with the target measured temperature allows for local sampling and detection of the temperature field prediction results. When the difference between the local predicted temperature and the target measured temperature is less than a deviation threshold, the temperature field prediction result is relatively accurate compared to the measured temperature and can be used as the temperature field reconstruction result for the entire target boiler. When the difference between the local predicted temperature and the target measured temperature is greater than or equal to the deviation threshold, the temperature field prediction result is not ideal. In this case, the temperature field prediction result is used as the initial temperature field, and the temperature field prediction model is called again to obtain a new temperature field prediction result. This is repeated until the difference between the extracted local predicted temperature and the target measured temperature is less than the deviation threshold. Once the accuracy of the temperature field prediction result is determined to be within the error range, the output of the temperature field prediction model is used as the temperature field reconstruction result for the target boiler.

[0152] In the embodiments of this application, the temperature field prediction model can be called multiple times. Therefore, the initial temperature field can be the local measured temperature of the target boiler, or it can be the temperature field prediction result when the difference between the local predicted temperature and the target measured temperature is greater than or equal to the deviation threshold.

[0153] Furthermore, the temperature field prediction model can also establish a time-series prediction model based on the model input parameters, predict the dynamic change trend of future model data parameters, predict the dynamic change of the target boiler temperature field in the future time period, identify possible combustion anomalies or local high temperature risks, and provide early warning or advance planning of prevention strategies.

[0154] As an example, we obtain measured datasets from multiple historical moments to form a historical dataset; based on the historical dataset and the time series prediction model, we obtain a predicted dataset; and input the predicted dataset into the temperature field prediction model to obtain the predicted temperature field reconstruction results.

[0155] As an example, a 300MW subcritical pressure, single intermediate reheat, single furnace, natural circulation, balanced ventilation, solid slag discharge, drum type, tangential coal-fired boiler is used as the target boiler. The boiler is pre-installed with a real-time furnace temperature measurement system based on high-temperature infrared transmitters and receivers, and 4x4=16 temperature data points are measured on the 60m and 75m sections of the furnace.

[0156] First, numerical simulation of the boiler furnace was performed using ANSYS Fluent 2025 and OpenFOAM V12. Geometric modeling and mesh generation were performed on the target boiler: ANSYS Fluent was used to build the boiler furnace model, employing a structured mesh with a total mesh size of 3 million and approximately 9 million edges. Figure 5 As shown.

[0157] The physical model is set as follows: turbulence model: RNG k-ε model; combustion model: non-premixed combustion model (EDM); radiative heat transfer model: P-1 radiation model; chemical reaction mechanism: based on the chemical reaction mechanism of pulverized coal combustion.

[0158] Boundary conditions were set as follows: primary air velocity: 10-15 m / s; secondary air velocity: 6-20 m / s; fuel flow rate: 4000 kcal pulverized coal, 100-200 t / h; water-cooled wall surface temperature: 200-300 degrees Celsius; superheater / reheater wall surface temperature: 400-560 degrees Celsius; solution and calculation: the SIMPLE algorithm and first-order scheme were used for pressure-velocity coupled solution; the time step was 0.001 s, the calculation was performed in 20,000 steps, and a single case could converge in about 6 hours to obtain the calculation results of the target boiler simulation data.

[0159] For the boundary conditions set above, the number of samples per case is 5, and the number of simulation cases is 3125. Each case stores 100 steps of intermediate and final convergence result data. Since there is no actual measured temperature data, noise is added to simulate the sensor installation location based on the convergence result data.

[0160] The temperature field prediction model is designed as follows: the unit model has 5 iterations, the node model and the edge model have 3 iterations each, the number of grid nodes is 3 million, and the number of edges is approximately 9 million.

[0161] The temperature field dataset of the target boiler was input into the temperature field prediction model. The CFD simulation calculation time was 6 hours; the inference time was 3-5 seconds using an ARMGNet 8xA100 inference server; the prediction accuracy was <3%.

[0162] In practical applications, the following beneficial effects exist:

[0163] Enhanced dynamic response capability: Enables second-level temperature prediction and rapid identification of abnormal combustion zones; Enhanced safety: Prevents furnace overheating and extends equipment lifespan; Energy saving optimization: Adjusts the air-coal ratio, reducing coal consumption by 1.5%.

[0164] This application first constructs a measured dataset based on the grid data and local measured temperatures of the target boiler. The measured dataset is then input into a temperature field prediction model to obtain the temperature field prediction results for the target boiler. Based on the target measured temperatures at preset locations and the temperature field prediction results, the temperature field of the target boiler is reconstructed. The temperature field prediction model enables autoregression of the unit model data, thereby correcting the data and obtaining more accurate temperature data while reflecting spatial relationships, thus correcting the initial temperature field and obtaining a relatively accurate temperature field. Finally, the temperature field prediction results are verified using a small amount of local measured temperatures. This demonstrates that high-precision reconstruction of the internal temperature field of the target boiler can be achieved using only a small amount of measured temperature data, providing theoretical support for combustion optimization, equipment safety management, and improved operating efficiency.

[0165] Although the steps in the above embodiments are described in the above order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not need to be executed in such order. They can be executed simultaneously (in parallel) or in reverse order. These simple changes are all within the protection scope of this application.

[0166] Please see Figure 6 The temperature field prediction model training system of the third embodiment of this application includes: a geometric model acquisition module 610, a temperature field data acquisition module 620, a unit model training module 630, and a temperature field prediction model generation module 640.

[0167] The geometric model acquisition module 610 is used to acquire the geometric model of the target boiler and to perform meshing on the geometric model, wherein the mesh of the geometric model corresponds to the mesh data used for positioning.

[0168] Temperature field data acquisition module 620 is used to acquire temperature field data of the target boiler based on the fluid numerical simulation model of the target boiler.

[0169] The unit model training module 630 is used to use the target dataset constructed from grid data and temperature field data as training samples for the unit model, and to train iteratively on the node temperature update model and the edge feature update model included in the unit model respectively.

[0170] The temperature field prediction model generation module 640 is used to synchronize the model parameters of the trained unit model to multiple layers to generate a temperature field prediction model composed of multiple layers, wherein the structure of the layers is the same as the structure of the unit model.

[0171] Please see Figure 7 The temperature field reconstruction system of the fourth embodiment of this application includes: a measured dataset construction module 710, a temperature field prediction result acquisition module 720, and a temperature field reconstruction module 730.

[0172] The measured dataset construction module 710 constructs a measured dataset based on the grid data and local measured temperatures of the target boiler.

[0173] The temperature field prediction result acquisition module 720 inputs the measured dataset into the temperature field prediction model to obtain the temperature field prediction result of the target boiler. The temperature field prediction model is trained by the above-mentioned temperature field prediction model training method.

[0174] The temperature field reconstruction module 730 reconstructs the temperature field of the target boiler based on the measured temperature of the target at a preset location and the predicted temperature field.

[0175] Please see Figure 8 The temperature field reconstruction system of the fifth embodiment of this application includes a sensor, a communication card, a data server, and an inference server.

[0176] In this embodiment of the application, the sensor is used to measure the furnace temperature of the target boiler.

[0177] Among them, furnace temperature sensors can measure the internal or wall temperature of the furnace directly or indirectly based on thermocouples, infrared, acoustic waves, etc.

[0178] The communication card is used to transmit the temperature data measured by the sensor to the data server.

[0179] Among them, the communication card is the communication card of the distributed control system (DCS).

[0180] The data server is used to receive temperature data; the data server is also used to construct the measured dataset; the data server is also used to transmit the measured dataset to the inference server; the data server is also used to receive the temperature field prediction results from the inference server, and reconstruct the temperature field of the target boiler based on the temperature field prediction results and the target measured temperature.

[0181] The inference server stores a temperature field prediction model, which is used to receive the measured dataset from the data server, execute the temperature field prediction model based on the measured dataset to obtain the temperature field prediction result, and return the temperature field prediction result to the data server.

[0182] Furthermore, the data server can also acquire temperature data alarm data for key areas based on the temperature field reconstruction results and transmit it through communication cards; the data server can also store the continuous data of the target boiler into a local database.

[0183] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0184] It should be noted that the boiler furnace temperature field reconstruction system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of this application can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of this application are only for distinguishing the various modules or steps and are not considered as an improper limitation of this application.

[0185] An electronic device according to the sixth embodiment of this application includes:

[0186] At least one processor; and

[0187] A memory communicatively connected to at least one of the processors; wherein,

[0188] The memory stores instructions that can be executed by the processor to implement the above-described boiler furnace temperature field reconstruction method.

[0189] A computer-readable storage medium according to a seventh embodiment of this application stores computer instructions that are executed by the computer to implement the above-described boiler furnace temperature field reconstruction method.

[0190] An eighth embodiment of this application provides a computer program product that, when run on an electronic device, causes the electronic device to execute the above-described boiler furnace temperature field reconstruction method.

[0191] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the electronic device and computer-readable storage medium described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0192] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0193] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0194] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0195] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0196] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0197] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.

Claims

1. A method for training a temperature field prediction model, characterized in that, The method comprises the following steps: acquire a geometric model of a target boiler, and perform meshing on the geometric model, wherein the mesh of the geometric model corresponds to mesh data for positioning, and the mesh data comprises mesh three-dimensional coordinates of each node in the mesh; acquire temperature field data of the target boiler based on a fluid numerical simulation model of the target boiler; construct a target data set of the mesh data and the temperature field data as a training sample of a unit model, and iteratively train a node temperature updating model and an edge feature updating model included in the unit model respectively; synchronize model parameters of the trained unit model to multiple layers to generate a temperature field prediction model composed of the multiple layers, wherein the structure of the layers is the same as that of the unit model; wherein the iteratively training the node temperature updating model and the edge feature updating model included in the unit model respectively comprises: acquire an initial node model and an initial edge model of the mesh, wherein the input data of the initial node model comprises an initial node temperature and an edge feature, and the output data of the initial node model is an updated node temperature, and the edge feature is an initial edge feature or an updated edge feature, the input data of the initial edge model comprises an initial edge feature and node features corresponding to nodes connected by the edge, and the output data is an updated edge feature, wherein the node features comprise mesh three-dimensional coordinates, measured temperature, initial temperature, furnace pressure and edge features corresponding to the nodes; use the updated node temperature of the initial node model at a first time as the initial node temperature at a second time, and iteratively train the initial node model to obtain the node temperature updating model; use the updated edge feature of the initial edge model at the first time as the initial edge feature at the second time, and iteratively train the initial edge model to obtain the edge feature updating model; acquire a node-edge critical matrix of the mesh, and couple the node-edge critical matrix, the node temperature updating model and the edge feature updating model to generate the unit model, wherein the node-edge critical matrix is used to represent the correspondence between the nodes and the edges of the mesh. 2.The temperature field prediction model training method of claim 1, wherein, In the case where the edge feature updating model converges, acquire the target edge feature output by the edge feature updating model as the edge feature in the input data of the node temperature updating model. 3.The temperature field prediction model training method of claim 1, wherein, The method comprises the following steps: acquire simulation data of combustion of multiple boiler furnaces based on the fluid numerical simulation model of the target boiler, wherein the simulation data comprises boiler furnace geometric parameters, mesh data and internal temperature field data of the boiler furnace, and the simulation data is used to simulate temperature field distribution under different furnace types and different operating conditions; perform data enhancement on the internal temperature field data of the boiler furnace to obtain enhanced temperature field data; splice the enhanced temperature field data and other data of the simulation data to generate the temperature field data of the target boiler. 4.The temperature field prediction model training method of claim 3, wherein, The method comprises the following steps: Based on fluid numerical modeling and simulation, a fluid, combustion and heat transfer model in any boiler furnace is constructed; The fluid, combustion and heat transfer model is solved and calculated to obtain a simulation calculation data set as simulation data of combustion in any boiler furnace. 5.The temperature field prediction model training method of claim 3, wherein, The data enhancement on the internal temperature field data of the boiler furnace is performed to obtain enhanced temperature field data, including: An intermediate iteration result of the simulation data in an iteration process is obtained as the enhanced temperature field data. 6.The temperature field prediction model training method of claim 5, wherein, The data enhancement on the internal temperature field data of the boiler furnace is performed to obtain enhanced temperature field data, and further including: An actual measured temperature of a random node position is obtained; The temperature at the random node position in the intermediate iteration result is replaced by the actual measured temperature to obtain the enhanced temperature field data.

7. A temperature field reconstruction method based on the temperature field prediction model training method of any one of claims 1-6, characterized in that, Including: Based on grid data and local actual measured temperature of a target boiler, an actual measured data set is constructed; The actual measured data set is input into a temperature field prediction model to obtain a temperature field prediction result of the target boiler; Based on a target actual measured temperature at a preset position and the temperature field prediction result, a temperature field of the target boiler is reconstructed.

8. The temperature field reconstruction method of claim 7, wherein, The temperature field reconstruction of the target boiler based on the target actual measured temperature at the preset position and the temperature field prediction result includes: A local prediction temperature of the preset position is extracted from the temperature field prediction result; In a case where a difference between the local prediction temperature and the target actual measured temperature is less than a deviation threshold, the temperature field prediction result is taken as a temperature field reconstruction result of the target boiler; In a case where the difference between the local prediction temperature and the target actual measured temperature is greater than or equal to the deviation threshold, the temperature field prediction result is input into the temperature field prediction model again for temperature field prediction until the difference between the local prediction temperature extracted from the temperature field prediction result and the target actual measured temperature is less than the deviation threshold, and an output result of the temperature field prediction model is taken as the temperature field reconstruction result of the target boiler.

9. The temperature field reconstruction method of claim 7, wherein, The method further includes: Actual measured data sets at multiple historical time points are obtained to form a historical data set; Based on the historical data set and a time series prediction model, a prediction data set is obtained; The prediction data set is input into the temperature field prediction model to obtain a prediction temperature field reconstruction result.

Images