Boiler heating surface overtemperature real-time monitoring method and system based on coupling model

Abstract

The invention discloses a boiler heating surface overtemperature real-time monitoring method and system based on a coupling model, and the monitoring method comprises the steps: firstly, building a deep learning model based on the historical operation condition of a boiler system with boiler operation data and heating surface temperature distribution as input and heating surface heat flow distribution as output; obtaining new boiler operation data in real time, and performing coupling calculation based on the constructed deep learning model and the dynamic simulation model so as to predict heat flow distribution and temperature distribution of the heating surface; and finally, judging the overtemperature condition of the boiler heating surface according to the predicted heat flow distribution and temperature distribution of the heating surface. According to the method, the deep learning model is introduced to replace a numerical simulation model, so that the deep learning model is coupled with the dynamic simulation model, rapid and accurate prediction of heat flow distribution and temperature distribution of the heating surface is realized through iterative calculation of the deep learning model and the dynamic simulation model, and the instantaneity, comprehensiveness and accuracy of overtemperature monitoring of the boiler heating surface are improved.

Description

Technical Field

[0001] The present invention relates to a method and system for real-time monitoring of overheating of boiler heating surfaces based on a coupling model, and belongs to the field of monitoring of power plant boiler systems. Background Art

[0002] Overheating of boiler heating surfaces is a serious problem in power plant operation and can cause various harms. When the temperature of the heating surface exceeds the designed temperature of the boiler pipe material, creep will gradually change the shape and size of the pipe material, making it unable to work properly. The thermal fatigue caused by overheating will generate micro-cracks inside the pipe material. With the passage of time and the repeated action of thermal stress, these cracks will continuously expand, reducing the strength and reliability of the pipe material. At the same time, excessive formation of pipe material oxide scale will not only reduce the heat transfer efficiency but also create conditions for corrosion to accelerate the damage of the pipe material, seriously shortening the service life of the pipe material and even possibly triggering safety accidents, posing a serious threat to personnel and equipment. From the perspective of operating costs, overheating of boiler heating surfaces is one of the main reasons for unplanned shutdowns of coal-fired boilers. Such unplanned shutdowns not only disrupt the normal power generation plan, affect the stability of power supply, but also cause huge economic pressure on enterprises.

[0003] The existing monitoring of boiler overheating mainly relies on the boiler distributed control system (DCS) equipped with thermocouples. The distributed control system collects various operating parameters including pressure, flow rate, temperature, etc. by installing sensors at various parts of the boiler, and processes and analyzes them through complex algorithms and logical judgments to reflect the overall operating state of the boiler. However, due to the usually large scale of the boiler and the high-temperature working environment, the data that can be collected by thermocouples is relatively limited. There are certain monitoring blind spots and obvious limitations in monitoring overheating of heating surfaces only through this distributed control system.

[0004] And a currently disclosed coupling calculation model for the wall temperature of boiler water walls accurately predicts the wall temperature distribution by coupling a three-dimensional CFD model describing the flow, combustion, and heat transfer processes of flue gas in the furnace and a one-dimensional hydrodynamic model describing the flow and heat transfer processes of the working medium in the water wall tubes on the basis of the monitoring data collected by the distributed control system. However, although this coupling calculation model has solved the prediction limitations of traditional numerical simulation methods, it still has the disadvantages of long calculation time and inability to cover all working conditions. Summary of the Invention

[0005] Objective of the Invention: Aiming at the problems existing in the prior art, the present invention provides a real-time monitoring method and system for overheating of boiler heating surfaces based on a coupling model. By introducing a deep learning model to replace the numerical simulation model and coupling it with the dynamic simulation model, rapid and accurate prediction of the heat flux distribution and temperature distribution of the heating surfaces is realized through the iterative calculation of the two, thereby improving the real-time performance, comprehensiveness, and accuracy of overheating monitoring of boiler heating surfaces.

[0006] Technical Solution: To achieve the above objective, the present invention provides a real-time monitoring method for overheating of boiler heating surfaces based on a coupling model, including the following steps:

[0007] S1. Based on the historical operating conditions of the boiler system, a deep learning model is constructed with boiler operation data and the temperature distribution of the heating surfaces as inputs and the heat flux distribution of the heating surfaces as outputs.

[0008] S2. New boiler operation data is obtained in real time, and coupling calculations are performed based on the constructed deep learning model and the dynamic simulation model to predict the heat flux distribution and temperature distribution of the heating surfaces.

[0009] S3. The overheating condition of the boiler heating surfaces is judged according to the predicted heat flux distribution and temperature distribution of the heating surfaces.

[0010] Specifically, the step S1 includes:

[0011] A1. Based on the historical operating conditions of the boiler system, a three-dimensional CFD model and a one-dimensional hydrodynamic model of the boiler system are respectively established through a numerical simulation model and a dynamic simulation model.

[0012] A2. The one-dimensional hydrodynamic model and the three-dimensional CFD model are coupled for calculation to predict the heat flux distribution and temperature distribution of the heating surfaces, thereby constructing a training data set for the deep learning model.

[0013] A3. Based on the training data set, a deep learning model is established with boiler operation data and the temperature distribution of the heating surfaces as inputs and the heat flux distribution of the heating surfaces as outputs.

[0014] Specifically, the dynamic simulation model is constructed based on the discrete element model. The one-dimensional hydrodynamic model is constructed by discretizing the heating surfaces, and each discrete unit consists of a heat flux control signal component, a heat flux component, a wall component, and a pipeline component.

[0015] Specifically, before establishing the deep learning model, preprocessing and proper orthogonal decomposition processing are performed on the training data set.

[0016] Specifically, the deep learning model adopts a U-Net structure, which includes an input layer, a convolutional layer, a pooling layer, an upsampling layer, and an output layer. The input layer is used to receive input parameters, the convolutional layer is used to extract parameter features, the pooling layer is used to reduce the resolution, the upsampling layer is used to restore the resolution and fuse feature information, and the output layer is used to predict the modal coefficients of each mode, thereby predicting the heat flux distribution of the heating surface.

[0017] Specifically, the loss function of the deep learning model is the mean square error, and the SGD and Adam optimization algorithms are used for optimization during the training process.

[0018] Specifically, the step S2 includes:

[0019] B1. Based on the new boiler operation data and the initial heating surface temperature distribution, the initial heating surface heat flux distribution is predicted by the deep learning model, and further substituted into the dynamic simulation model to calculate the updated heating surface temperature distribution;

[0020] B2. Substitute the updated heating surface temperature distribution into the deep learning model for iterative calculation, and continuously perform real-time coupling calculation until the error between the heating surface temperature distributions before and after the update is within the set range, to obtain the finally predicted heating surface heat flux distribution and temperature distribution.

[0021] Specifically, the step S3 includes:

[0022] Discretize the grid of the heating surface. According to the heating surface temperature distribution predicted by the dynamic simulation model, judge whether the heating surface temperature is over-temperature grid by grid through threshold analysis, thereby statistically calculating the area of the heating surface area where the temperature is higher than the over-temperature threshold, and further judging the over-temperature situation of the boiler heating surface.

[0023] Specifically, the output of the deep learning model also includes the flue gas temperature distribution of the three-dimensional combustion field. Furthermore, based on the three-dimensional flue gas temperature predicted by the deep learning model, the flame center position is determined by finding the high-temperature area in the temperature field, thereby assisting in judging the over-temperature situation of the boiler heating surface.

[0024] In addition, the present invention also provides a real-time monitoring system for boiler heating surface over-temperature based on a coupling model, including a collection module and a processing module. The processing module uses the above-mentioned real-time monitoring method for boiler heating surface over-temperature to perform real-time monitoring according to the boiler operation data collected by the collection module in real time.

[0025] Beneficial effects: By introducing a deep learning model to replace the numerical simulation model and coupling it with the dynamic simulation model, the present invention realizes the rapid and accurate prediction of the heat flux distribution and temperature distribution of the heating surface through the iterative calculation of the two, thereby improving the real-time performance, comprehensiveness, and accuracy of the boiler heating surface over-temperature monitoring.

[0026] When constructing the training data set, on the one hand, a numerical simulation model is used to accurately simulate the combustion and radiative heat transfer in the boiler furnace, and on the other hand, a dynamic simulation model is used to carefully reflect the steam-water and heat transfer conditions of the working medium in the heating surface. The two complement each other, and finally achieve an accurate prediction of the wall temperature distribution of the heating surface for subsequent deep learning training, thereby improving the prediction accuracy of the deep learning model.

[0027] In addition, the deep learning model is trained with a large amount of data, and can quickly predict the modal coefficients based on a variety of boiler operation parameters, reconstruct a fine multi-physical field, achieve a rapid prediction of the distribution of multiple parameters in the furnace and the wall heat flux distribution, and can early warn of over-temperature conditions, enhancing the timeliness of over-temperature monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic flow chart of the method for real-time monitoring of over-temperature of the boiler heating surface in the embodiment of the present invention;

[0029] Figure 2 is a schematic flow chart of the coupling model in the embodiment of the present invention;

[0030] Figure 3 is a schematic diagram of the construction of the discrete model in the embodiment of the present invention;

[0031] Figure 4 is a schematic diagram of the construction of the discrete model of the spiral water wall in the embodiment of the present invention;

[0032] Figure 5 is a schematic diagram of the composition of the real-time monitoring system for over-temperature of the boiler heating surface in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention.

[0034] Referring to Figure 1 and Figure 2 , the present invention provides a method for real-time monitoring of over-temperature of a boiler heating surface based on a coupling model, including the following steps:

[0035] S1. Based on the historical operating conditions of the boiler system, a deep learning model is constructed with the boiler operation data and the heating surface temperature distribution as inputs and the heating surface heat flux distribution as the output;

[0036] In this embodiment, a 600MW supercritical swirl opposed fired boiler is selected as the simulation object, and 300 sets of the most frequently occurring stable operating conditions are selected from the historical data of the power plant DCS system. These conditions can best reflect the performance characteristics of the boiler during normal operation and provide a representative operating condition basis for the model.

[0037] 1.1. Establish a numerical simulation model of the boiler system;

[0038] a. Numerical simulation platform:

[0039] Ansys Fluent software is used to carry out three-dimensional CFD simulations of the flow, combustion, and heat transfer processes on the flue gas side of the boiler. This software has powerful functions and wide applications in the field of fluid mechanics simulation and can effectively handle complex physical phenomena on the flue gas side of the boiler. By simulating the flow, combustion, and heat transfer processes on the flue gas side, the heat flux distribution on the heating surface can be obtained. In addition, data such as the flue gas temperature and gas components in the entire three-dimensional combustion field can also be obtained.

[0040] b. Turbulent flow model:

[0041] The Realizable k-ε model is adopted, and its k equation and ε equation are:

[0042]

[0043] where η = Sk / ε, S = (2S ij S ij ) 1 / 2 , and the constant values are C μ = 0.0845, σ k = 0.7194, σ ε = 0.7194, C 1ε = 1.42, C 2ε = 1.68, η 0 = 4.38, β = 0.012. This model can better adapt to the turbulent flow characteristics in the boiler and accurately simulate the influence of turbulence on flow and heat transfer.

[0044] c. Radiation heat transfer model:

[0045] The discrete ordinates method (DO) model is used to calculate radiation heat transfer, and the weighted sum of gray gases model (WSGG) is used to calculate the flue gas absorption coefficient. This model can take into account the complex interaction between gas and particle radiation when dealing with radiation heat transfer, improve the accuracy of radiation heat transfer calculation, and is especially suitable for the complex combustion environment in the boiler.

[0046] d. Particle motion model:

[0047] Track the path of pulverized coal particles in the Lagrangian coordinate system with a random trajectory model. This model can describe in detail the movement trajectory of pulverized coal particles in a complex flow field, considering various forces acting on the particles (such as drag force, gravity, turbulent pulsation, etc.), providing a basis for accurately simulating the pulverized coal combustion process.

[0048] e. Pulverized coal parameters:

[0049] Set the average diameter of pulverized coal to 57 μm and the uniformity index to 1.13. Use the composite volatile decomposition (cpd) model to calculate the volatile release rate (the total volatile release under high-temperature pyrolysis conditions is 35.3% on a dry basis). These parameters are based on the study of the actual combustion characteristics of pulverized coal and can reflect the pyrolysis and combustion behavior of pulverized coal in the boiler.

[0050] f. Reaction rate model:

[0051] Use the diffusion and kinetic control rate model to calculate the oxidation reaction rate on the surface of pulverized coal. Based on the Baum and Street and Field models integrated in ANYSY FLUENT, this model comprehensively considers the influence of diffusion and kinetic factors on the reaction rate, making the calculation of the pulverized coal combustion reaction rate more in line with the actual situation.

[0052] g. Gas-phase chemical reaction simulation:

[0053] Use the improved WD-1 mechanism to simulate gas-phase chemical reactions. This mechanism can more accurately describe the process and rate of various chemical reactions in the gas phase, considering the interaction between chemical reactions and turbulence, which is crucial for simulating the change of gas-phase components and energy release during the combustion process.

[0054] h. Boundary conditions:

[0055] Set the mass flow inlet boundary conditions for the primary air, secondary air, overfire air, and side overfire air. Define the swirl angle of the burner using the local cylindrical coordinate system. The inlet air volume and air temperature are set according to the boiler design specification and the simulation results of the primary air structure to ensure that the inlet conditions conform to the actual operating conditions. The outflow boundary condition is used at the outlet to allow the fluid to flow out of the computational domain freely. The furnace wall is simulated using the standard wall equation (the velocity at the fluid-wall contact is zero) to simulate the wall adhesion phenomenon, and the wall heat exchange is treated with the convective boundary condition, and the wall water vapor temperature and radiation emissivity are given to simulate the heat energy exchange. The diameter of pulverized coal particles is defined by the R-R formula (the distribution index n is 1.13, the average diameter is 57 μm, and the particle size range is 3 - 250 microns), and this formula can accurately describe the particle size distribution characteristics of pulverized coal particles.

[0056] 1.2. Establish a dynamic simulation model of the boiler system;

[0057] A. Discrete element model:

[0058] Reference Figure 3 Figure 3 , the heating surface (such as the spiral water wall and the platen superheater) is discretized. Each discrete unit consists of a heat flux control signal component, a heat flux component, a wall component, and a pipeline component. When constructing the discrete unit model, the heat transfer characteristics in the pipeline under supercritical conditions are fully considered, and different correlation formulas are selected for heat transfer calculation. The selection of these heat transfer correlation formulas is based on experimental research and data analysis of the heat transfer characteristics of different tube types under supercritical pressure, and can more accurately reflect the heat transfer situation of the working fluid in the tube.

[0059] B. Pressure loss model:

[0060] The Darcy formula is used for pressure loss calculation:

[0061]

[0062] where h f is the head loss along the way, f is the friction resistance coefficient, l is the pipe length, d is the pipe diameter, v is the average flow velocity of the fluid in the pipe, and g is the acceleration due to gravity.

[0063] For single-phase fluids with a smooth surface, the Blasius formula is used to calculate the friction resistance coefficient f:

[0064]

[0065] where Re f is the Reynolds number with the working fluid temperature as the qualitative temperature, G is the mass flow rate, and μ f is the dynamic viscosity of the working fluid.

[0066] For single-phase fluids with a rough surface, the Colebrook-White formula is used to calculate the friction resistance coefficient f:

[0067]

[0068] where ε is the pipe wall roughness and D is the pipe diameter.

[0069] The pressure loss model selects the Colebrook-White formula in Modelica to calculate the pressure loss in the pipe. The model is driven by pressure and checks the total flow rate by adjusting the average roughness in the pipe to ensure the accuracy of the pressure loss calculation. This calculation method considers the energy loss caused by factors such as friction and local resistance when the fluid flows in the pipeline, which is crucial for accurately simulating the flow state of the working fluid.

[0070] C. Discrete structure division and overall model construction:

[0071] The heating surface is divided into discrete structures according to the number and location of the wall temperature measurement points installed on site.

[0072] Refer to Figure 4 Figure 4 , taking the spiral water wall as an example, there are 8 thermocouples on each of the front and rear walls, and 5 thermocouples on each of the left and right walls. To reflect the differences between different measuring points, the front and rear walls are discretized transversely by 10 units, and the left and right walls are discretized transversely by 7 units. The total height from the cold ash hopper outlet to the spiral water wall outlet is about 31m, and it is discretized longitudinally by 40 units, totaling 1360 units. Based on the discrete element model, a discrete model of the spiral water wall is built according to this discrete scheme, and a pipeline model that can be discretized in one dimension (the number of discretizations is 40) is used.

[0073] For the platen superheater, there are two rows of tube screens along the furnace depth direction, and each row of tube screens has 15 pieces along the furnace width direction, totaling 30 screens. Each screen is composed of 22 tube coils. The inlet section and the outlet section of each screen are discretized into 3 units respectively, and the bottom of the screen is represented by 1 unit, totaling 210 units. Then, they are spliced to form a one-dimensional hydrodynamic model of the platen superheater. This way of discrete structure division can reflect the working fluid flow and heat transfer conditions at different positions of the heating surface in detail, providing a basis for accurately predicting the wall temperature distribution.

[0074] 1.3. Produce the training dataset for the deep learning model;

[0075] Couple the one-dimensional hydrodynamic model with the three-dimensional CFD model to predict the heat flux and wall temperature distribution of the heating surface, and realize data interaction through Python during the calculation process. First, calculate with the uniform working fluid temperature as the wall boundary condition of the three-dimensional CFD model. After obtaining the heat flux density distribution, return the heat flow corresponding to each control volume to the Dymola model according to the coordinates, and further calculate the temperature of the working fluid in the pipe in the Dymola model.

[0076] Then, fit the calculated working fluid temperature distribution data into a polynomial function that changes along the coordinates, and transfer the temperature distribution function back to the CFD model on the flue gas side through the udf file to update the wall boundary for iterative calculation until the calculation deviation value meets the convergence condition (the temperature difference of the working fluid is less than 1K). During the iterative calculation process, continuously adjust the calculation results of the working fluid temperature and the heat flux density to make them match each other, and finally realize the accurate prediction of the wall temperature distribution of the heating surface. This data interaction and iterative calculation method fully considers the mutual influence between the combustion side and the working fluid side, improving the accuracy and reliability of the wall temperature prediction. Finally, use Python code to process the calculated results into a multi-channel three-dimensional.npy dataset file for subsequent deep learning training.

[0077] When calculating the temperature of the working fluid in the pipe, the convective heat transfer equation between the working fluid and the inner wall surface needs to be considered:

[0078] Φ = αA(t w - t f ),

[0079] Among them, Φ is the heat transfer amount, α is the convective heat transfer coefficient, A is the effective heat transfer area, and t ω is the inner wall surface temperature, and t f is the working medium temperature, and the tube wall heat conduction equation:

[0080]

[0081] Among them, λ is the tube wall thermal conductivity, l is the tube length, and r o is the outer tube wall radius, and r ω is the inner tube wall radius, and t o is the outer wall temperature.

[0082] 1.4. Establish a deep learning model for boiler three-dimensional field prediction;

[0083] 1) Preprocessing of the training dataset:

[0084] In order to comprehensively cover various operating states of the boiler, extensive historical operating data of power station boilers are collected. After in-depth analysis of the boiler operating parameter range, 300 sets of stable operating conditions are carefully designed for iterative calculation, thereby generating the corresponding FOD database. These 300 sets of operating conditions can more comprehensively reflect the operating conditions of the boiler under different loads, coal types, wind speeds, etc.

[0085] In the preprocessing stage, first, data cleaning is performed to remove outliers and incorrect data to ensure the accuracy of the data. Then, normalization processing is carried out to map the values of different parameters to a specific interval. For example, the data is normalized to the [0, 1] interval, and the calculation formula is:

[0086]

[0087] Among them, x is the original data, and x min and x max are the minimum and maximum values of this parameter, respectively.

[0088] Furthermore, the resolution can be reduced by downsampling to reduce the data volume without losing key information and reduce the computational cost. Dimensionality reduction operations can also be performed, such as using methods like principal component analysis (PCA) to reduce the feature dimensions of the data. At the same time, according to the distribution characteristics of the training dataset, data augmentation techniques, such as random rotation, flipping, scaling, etc., are used to expand the number of samples and improve the generalization ability and robustness of the model to different operating conditions.

[0089] 2) Proper Orthogonal Decomposition (POD) processing:

[0090] Introduce the POD decomposition strategy for the original multi-physical field data:

[0091] u(x, t),

[0092] where \(x\) is the spatial coordinate and \(t\) is the time.

[0093] By solving the eigenvalue problem:

[0094]

[0095] where \(C(x,y)\) is the correlation function, \(\Omega\) is the computational domain, \(\lambda\) k is the eigenvalue, is the eigenfunction, i.e., the spatial mode function, thus obtaining the spatial mode function.

[0096] Then calculate the modal coefficients:

[0097]

[0098] Thus, the high-dimensional and complex multi-physical field data is decomposed into:

[0099]

[0100] In practical applications, the main modes are selected for combination according to principles such as energy contribution, etc., to achieve effective dimensionality reduction of the data, retain the information that has a greater impact on the combustion process, reduce the computational amount, and ensure the accuracy of the model at the same time.

[0101] 3) Construction and training of the deep learning model:

[0102] Construct a model structure based on the U-Net deep learning network: Take various boiler operating parameters (such as coal type, burnout air velocity, load, etc.) and the temperature distribution of the heating surface as the input conditions of the model. The input layer is designed according to the quantity and type of the input parameters; the middle layer of the network adopts the typical structure of U-Net, including convolutional layers, pooling layers, upsampling layers, etc. The convolutional layers are used to extract features, the pooling layers reduce the resolution, and the upsampling layers restore the resolution and fuse the feature information of different levels; the output layer predicts the modal coefficients of each mode.

[0103] Define the loss function as the mean square error (MSE). Let the predicted modal coefficient be the true modal coefficient be \(a\) k , then the MSE formula is:

[0104]

[0105] where \(N\) is the number of samples.

[0106] During the training process, use the SGD (stochastic gradient descent) and Adam optimization algorithms. The SGD algorithm minimizes the loss function by continuously updating the parameters, and the update formula is:

[0107]

[0108] where η is the learning rate, is the gradient of the loss function with respect to the parameters.

[0109] The Adam optimization algorithm combines the ideas of momentum method and adaptive learning rate, uses different learning rates for different parameters, and can converge faster during training. The best configuration of the model is obtained through K-fold cross-validation (divide the dataset into K parts, use K - 1 parts as the training set each time, and the remaining 1 part as the validation set, repeat K times, and finally take the average performance index) and Bayesian hyperparameter tuning (using Bayesian statistical methods to estimate the posterior distribution of hyperparameters, so as to select the optimal hyperparameters), improving the prediction accuracy and generalization ability of the model.

[0110] S2. Obtain new boiler operation data, and perform coupled calculations based on the constructed deep learning model and dynamic simulation model, thereby predicting the heat flux distribution and wall temperature distribution of the heating surface.

[0111] In practical applications, coupled calculations are performed based on the constructed deep learning model and dynamic simulation model to predict the heat flux distribution and wall temperature distribution of the heating surface, and data interaction is realized through Python during the calculation process. First, when key parameters such as the operating load, coal type, mill status, temperature, and wind speed of the boiler are received as inputs, the calculation is performed with the uniform working fluid temperature as the initial heating surface temperature distribution. The modal coefficients of each mode are predicted through the deep learning model, and then these predicted modal coefficients are multiplied by the spatial matrix obtained in advance through POD decomposition:

[0112]

[0113] where K is the number of main modes selected, thereby reconstructing the complete refined multi-physical field, and thus realizing the rapid prediction of the concentration distributions of components such as furnace speed, temperature, oxygen, carbon monoxide, and carbon dioxide under the corresponding working conditions, as well as the heat flux distribution on the boiler wall surface.

[0114] Subsequently, use Python code to process the predicted wall heat flux results of the deep learning model according to the discrete division scheme pre-constructed by the dynamic simulation model, calculate the average heat flux value in each discrete region, and thus accurately input the average heat flux values of each discrete region into the dynamic simulation model. Then, according to the preset real-time simulation step size, after the dynamic simulation model performs a period of simulation calculation, the wall temperature result data of each discrete region in the dynamic simulation model are output, and these output wall temperature data will be passed back to the deep learning model as the updated wall temperature conditions.

[0115] Finally, the deep learning model integrates these new wall temperature conditions and real-time operating conditions parameters, and starts the prediction process again to predict the updated wall heat flux data. Through such iterative cycles and continuous real-time coupling calculations, the deep learning model and the dynamic simulation model can interact with each other and correct each other, thus more accurately mapping the real operating state of the boiler heating surface and providing more accurate and reliable data support for the overtemperature monitoring operation.

[0116] With the excellent parallel acceleration ability of the GPU, the above coupling process can achieve fast calculations in milliseconds. Based on these prediction results, the operator can timely understand the internal combustion state of the boiler and optimize the boiler operation, such as adjusting parameters such as the air distribution and coal feeding amount of the burners, so as to improve the combustion efficiency, reduce pollutant emissions, and ensure the safe and stable operation of the boiler.

[0117] S3. Judge the overtemperature situation of the boiler heating surface according to the predicted heat flux distribution and wall temperature distribution of the heating surface.

[0118] According to the data results obtained by coupling the above models, a visual interface for monitoring wall overtemperature is established to realize real-time monitoring of the state of the boiler heating surface. First, establish a data connection interface with the temperature monitoring system of the boiler heating surface to obtain the temperature data of each part of the heating surface pipeline in real time. Then, set the overtemperature threshold. When the received temperature data exceeds this threshold, dynamically adjust the material properties of the corresponding pipeline part. For example, modify the material color of the overtemperature pipeline part to a prominent red, and at the same time increase its transparency to make the overtemperature area more intuitively displayed; and a flashing effect can be added to attract the user's attention and timely prompt the operator of the overtemperature situation of the boiler heating surface pipeline.

[0119] In addition, according to the data results obtained by coupling the above models, some key parameters during the boiler operation are further calculated as the basis for judging the overtemperature of the heating surface. These key parameters include:

[0120] ① Calculation of the flame center position:

[0121] According to the temperature field data predicted by the deep learning model, determine the flame center position by finding the high-temperature area in the temperature field. Adopt a method based on the temperature threshold, set a relatively high temperature threshold, regard the area with a temperature higher than this threshold as the flame area, and then determine the flame center position by calculating the geometric center or mass center of this area. Let the coordinates of each point in the flame area be (x i , y i , z i ), and the temperature be T i , then the flame center position (x c , y c , z c) It can be calculated by using the weighted average method:

[0122]

[0123] ② Calculation of flue gas temperature deviation:

[0124] Among the flue gas temperature data output by the deep learning model, select multiple representative monitoring points at different positions at the furnace outlet and different cross-sections of the flue duct, calculate the temperature differences between these monitoring points, and use this as a measure of the flue gas temperature deviation. Let the temperature of monitoring point A be T A , and the temperature of monitoring point B be T B , then the flue gas temperature deviation ΔT = T A - T B .

[0125] ③ Calculation of the over-temperature area of the heating surface:

[0126] According to the temperature distribution data of the heating surface calculated by the dynamic simulation model, determine the over-temperature threshold (this threshold takes the allowable temperature of the heating surface material), and then count the area of the heating surface area where the temperature is higher than the over-temperature threshold. Specifically: according to the discrete grid division of the heating surface during modeling, judge grid by grid whether the temperature is over-temperature, and then accumulate the area of the over-temperature grids to calculate the over-temperature area. Let the total number of grids be N, the number of over-temperature grids be n, and the area of each grid be S, then the over-temperature area A of the heating surface = nS.

[0127] Referring to Figure 5 , the present invention also provides a real-time monitoring system for over-temperature of boiler heating surfaces based on a coupling model, including an acquisition module (such as a DCS system, sensor measurement points, and actuators, etc.) and a processing module (such as an over-temperature monitoring server for heating surfaces). Among them, the processing module uses the above-mentioned real-time monitoring method for over-temperature of boiler heating surfaces to perform real-time monitoring according to the boiler operation data collected by the acquisition module in real time.

[0128] In the present invention, the coupling system of the boiler dynamic simulation model, deep learning model, and numerical simulation model works together, thereby realizing the rapid and accurate prediction of the heat flux distribution and temperature distribution of the heating surface, effectively improving the real-time performance, comprehensiveness, and accuracy of over-temperature monitoring of the boiler heating surface.

[0129] In addition, the visualization interface can intuitively display the temperature distribution of the heating surface, and the over-temperature area is prompted in a prominent manner, which is convenient for operators to grasp the state of the heating surface in real time; based on the data of each model, key parameters such as the flame center, flue gas temperature deviation, and over-temperature area are calculated, providing a more accurate basis for judging over-temperature, helping operators to deeply understand the internal operation state of the boiler, enabling them to optimize the burner air distribution and adjust the coal feeding amount in a timely manner, improving the combustion efficiency, reducing pollutant emissions, ensuring the safe and stable operation of the boiler, reducing unplanned shutdowns, enhancing the stability of the power supply system, and reducing enterprise costs.

[0130] The above specific embodiments only describe the preferred embodiments of the present invention, rather than limiting the protection scope of the present invention. Without departing from the design concept and spirit scope of the present invention, various deformations, substitutions and improvements made by those of ordinary skill in the art to the technical solutions of the present invention according to the written description and drawings provided by the present invention shall fall within the protection scope of the present invention.

Claims

1. A real-time monitoring method for overtemperature of boiler heating surface based on coupling model, characterized in that: The following steps are involved: S1. Based on the historical operating conditions of the boiler system, a deep learning model is constructed with boiler operating data and heating surface temperature distribution as input and heating surface heat flow distribution as output; S2. Acquire new boiler operation data in real time, and perform coupled calculation based on the built deep learning model and dynamic simulation model to predict the heat flux distribution and temperature distribution of the heating surface; S3. Determine the over-temperature condition of the boiler heating surface based on the predicted heat flux distribution and temperature distribution of the heating surface.

2. The method for real-time monitoring of overtemperature of boiler heating surface according to claim 1, characterized in that: The step S1 comprises: A1. Based on the historical operating conditions of the boiler system, a three-dimensional CFD model and a one-dimensional hydrodynamic model of the boiler system are established through numerical simulation models and dynamic simulation models respectively; A2. Couple the one-dimensional hydrodynamic model with the three-dimensional CFD model to predict the heat flux distribution and temperature distribution of the heated surface, thereby constructing a training data set for the deep learning model; A3. Using boiler operation data and heating surface temperature distribution as input and heating surface heat flux distribution as output, a deep learning model is established based on the training data set.

3. The real-time monitoring method for overtemperature of boiler heating surface according to claim 2 is characterized in that: The dynamic simulation model is constructed based on a discrete unit model, and a one-dimensional hydrodynamic model is constructed by discretizing the heated surface. Each discrete unit consists of a thermal flow control signal component, a thermal flow component, a wall component and a pipeline component.

4. The real-time monitoring method for overtemperature of boiler heating surface according to claim 2, characterized in that: Before building the deep learning model, the training dataset is preprocessed and subjected to eigenorthogonal decomposition.

5. The method for real-time monitoring of overtemperature of boiler heating surface according to claim 4, characterized in that: The deep learning model adopts a U-Net structure, which includes an input layer, a convolution layer, a pooling layer, an upsampling layer and an output layer, wherein the input layer is used to receive input parameters, the convolution layer is used to extract parameter features, the pooling layer is used to reduce the resolution, the upsampling layer is used to restore the resolution and fuse feature information, and the output layer is used to predict the modal coefficients of each mode, thereby predicting the heat flux distribution of the heated surface.

6. The method for real-time monitoring of overtemperature of boiler heating surface according to claim 5, characterized in that: The loss function of the deep learning model is mean square error, and SGD and Adam optimization algorithms are used for optimization during the training process.

7. The method for real-time monitoring of overtemperature of boiler heating surface according to claim 1, characterized in that: The step S2 comprises: B1. Based on the new boiler operation data and the initial heating surface temperature distribution, the initial heating surface heat flux distribution is predicted by the deep learning model, and then substituted into the dynamic simulation model to calculate the updated heating surface temperature distribution; B2. Substitute the updated temperature distribution of the heated surface into the deep learning model for cyclic iteration, and continue to perform real-time coupling calculations until the temperature distribution error of the heated surface before and after the update is within the set range, and obtain the final predicted heat flux distribution and temperature distribution of the heated surface.

8. The method for real-time monitoring of overtemperature of boiler heating surface according to claim 1, characterized in that: The step S3 comprises: The heating surface is divided into discrete grids. According to the temperature distribution of the heating surface predicted by the dynamic simulation model, the threshold analysis is used to determine whether the heating surface temperature is over-temperature. The area of ​​the heating surface with a temperature higher than the over-temperature threshold is statistically analyzed to determine the over-temperature situation of the boiler heating surface.

9. The method for real-time monitoring of overtemperature of boiler heating surface according to claim 8, characterized in that: The output of the deep learning model also includes the flue gas temperature distribution of the three-dimensional combustion field. Based on the three-dimensional flue gas temperature predicted by the deep learning model, the center position of the flame is determined by finding the high-temperature area in the temperature field, thereby assisting in judging the over-temperature condition of the boiler heating surface.

10. A real-time monitoring system for overtemperature of boiler heating surface based on coupling model, characterized in that: It comprises a collection module and a processing module, wherein the processing module performs real-time monitoring based on the boiler operation data collected in real time by the collection module using the boiler heating surface overtemperature real-time monitoring method described in any one of claims 1 to 9.

Images