Coal-fired boiler temperature field digital twinborn model construction method

Abstract

The invention discloses a coal-fired boiler temperature field digital twinborn model construction method, and relates to the technical field of boiler temperature detection, and the method comprises the steps: obtaining temperature data and input working condition parameters of a coal-fired boiler under multi-working-condition numerical simulation; carrying out data expansion on the temperature data by adopting a measurement driving type data enhancement method to obtain multi-working-condition temperature field data; on the basis of the multi-working-condition temperature field data, a mode and a mode coefficient are obtained through an intrinsic orthogonal decomposition method; based on the input working condition parameters and the modal coefficients, an improved online least square support vector machine algorithm is adopted to construct a prediction model; predicting each order of current modal coefficient corresponding to the current input working condition parameter through the prediction model; and constructing a temperature field digital twinborn model of the coal-fired boiler based on the modality and the current modal coefficient of each order. According to the method, rapid and comprehensive prediction of the temperature field information of the coal-fired boiler under variable working conditions can be realized.

Description

Technical Field

[0001] The present application relates to the technical field of boiler temperature detection, and in particular to a method for constructing a digital twin model of a temperature field of a coal-fired boiler. Background Art

[0002] In the new round of energy revolution, renewable energy power generation represented by wind power generation and photovoltaic power generation has gradually emerged, but due to its intermittent, random and volatile characteristics, it is more difficult to regulate the power system, and the balance and safety issues of the power system are more prominent. It will take a certain transition period for renewable energy power generation to replace thermal power generation. In the future, the main force of power energy supply will still come from thermal power generation, and thermal power generation is an important area for carbon emission reduction. Therefore, how to achieve clean, low-carbon, safe and efficient operation of thermal power generation is a challenge that society needs to face urgently.

[0003] For the supply side of the thermal power industry, coal-fired boilers are the main source of carbon emissions. Stable control of the furnace combustion system can effectively reduce carbon emissions. The pulverized coal combustion process inside the boiler is an extremely complex reaction process, involving multiple physical and chemical changes such as combustion, heat transfer, and flow, which are affected by many factors. Temperature, as an important parameter in the combustion process, characterizes the state of energy conversion and transmission during the combustion process. Therefore, real-time monitoring of the furnace temperature field has very important practical application value. In the traditional furnace combustion monitoring system, furnace temperature monitoring can be divided into contact temperature measurement technology and non-contact temperature measurement technology. However, due to the harsh site conditions of power station boilers, expensive measurement equipment, and limited measurement information, it is difficult to grasp the combustion state inside the furnace. With the continuous development of computational fluid dynamics (CFD) technology and computer technology, although the CFD numerical simulation method can accurately reflect the operation of the boiler, especially in the flow, combustion, and pollutant release in the furnace, and visualize the simulation results, it is not affected by factors such as the actual site environment, but it is still subject to lengthy calculation time, which makes this method have certain limitations when applied in real-time regulation and operation. Summary of the invention

[0004] The purpose of this application is to provide a method for constructing a digital twin model of the temperature field of a coal-fired boiler, which can realize rapid and comprehensive prediction of the temperature field information of the coal-fired boiler under variable operating conditions.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] The present application provides a method for constructing a digital twin model of a temperature field of a coal-fired boiler, comprising:

[0007] Acquire temperature data and input operating condition parameters of a coal-fired boiler under multi-operating condition numerical simulation; the input operating condition parameters include the amount of coal burned by burners on each layer of the coal-fired boiler, the primary air volume of burners on each layer, the secondary air volume of burners on each layer, and the burnt air volume of burnt air nozzles on each layer;

[0008] The temperature data is expanded using a metric-driven data enhancement method to obtain temperature field data under multiple working conditions.

[0009] Based on the temperature field data of multiple working conditions, the modes and modal coefficients are obtained using the eigenorthogonal decomposition method;

[0010] Based on the input operating parameters and modal coefficients, the prediction model is constructed using an improved online least squares support vector machine algorithm;

[0011] The current modal coefficients of each order corresponding to the current input working condition parameters are predicted by the prediction model;

[0012] A temperature field digital twin model of a coal-fired boiler is constructed based on modes and current modal coefficients of each order; the temperature field digital twin model is used to predict the temperature field.

[0013] Optionally, the method of using a metric-driven data enhancement method to expand the temperature data to obtain multi-condition temperature field data specifically includes:

[0014] Based on the wth temperature data x w and the w+1th temperature data x w+1 , using the formula Calculate x w and x w+1 The difference measurement result d(x w ,x w+1 ), where n is the total dimension of the temperature data; is the wth temperature data x w The v dimension; is the w+1th temperature data x w+1 The v dimension;

[0015] According to the difference measurement result d(x w ,x w+1 ), through the formula Determine the data sparse interval and perform data expansion; where W is the total number of temperature data; δ s is the data sparse interval threshold.

[0016] Optionally, after the temperature data is expanded using the metric-driven data enhancement method, the method further includes:

[0017] After data expansion is completed, if the expanded data x nw Satisfies the formula max(d(xw ,x nw ),d(x nw ,x w+1 ))<d(x w ,x w+1 ), then expand the data x nw Valid; otherwise, expand the data x nw Invalid, cancel data expansion; where d(x w ,x nw ) is x w and x nw The difference measurement results between nw ,x w+1 ) is x nw and x w+1 The difference measurement results between .

[0018] Optionally, the method of obtaining the modes and modal coefficients based on the temperature field data of multiple working conditions by using the intrinsic orthogonal decomposition method specifically includes:

[0019] Based on the temperature field data of multiple working conditions, the intrinsic orthogonal decomposition method is used through the formula Calculate the i-th mode By formula Calculate the i-th mode The corresponding modal coefficient θ i ; Where N is the total number of modes; is the pulsation matrix; i and β i are the eigenvalues ​​and eigenvectors of the correlation matrix R,

[0020] Optionally, the prediction model is constructed based on the input operating condition parameters and modal coefficients by using an improved online least squares support vector machine algorithm, specifically including:

[0021] Based on the input operating parameters and modal coefficients, with 23-dimensional input operating parameters as the prediction model input and 1-dimensional modal coefficient as the prediction model output, the improved online least squares support vector machine algorithm is used to build and train the prediction model.

[0022] Optionally, the decision function of the improved online least squares support vector machine algorithm is Among them, α k is the Lagrange multiplier corresponding to the kth input operating condition parameter; K(x,x k ) is the kernel function; b is the deviation; l is the total number of input operating condition parameters.

[0023] Optionally, the building of a digital twin model of the temperature field of a coal-fired boiler based on the modes and current modal coefficients of each order specifically includes:

[0024] Based on the mode, the number of main modes is determined using the formula Get the reconstructed temperature field snapshot matrix U"; where, is the sth main mode; θ' is the current modal coefficient of each order; M is the total number of main modes; is the mean column vector;

[0025] A digital twin model of the temperature field of a coal-fired boiler is constructed based on the reconstructed temperature field snapshot matrix.

[0026] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0027] The present application provides a method for constructing a digital twin model of the temperature field of a coal-fired boiler, which mainly involves a method for constructing a digital twin model of the temperature field of a coal-fired boiler based on metric drive and model reduction. The temperature data is expanded by using a metric-driven data enhancement method, so that the acquired multi-operating condition temperature field data can cover multiple operating conditions more comprehensively, and improve data quality and reduce data redundancy while minimizing the number of CFD simulations. At the same time, in order to reduce the computational time required to obtain temperature field information, the present application uses an improved online least squares support vector machine algorithm to construct a prediction model, thereby obtaining a digital twin model of the temperature field of a coal-fired boiler, and uses a model reduction method to reduce the order of the prediction model, thereby realizing a rapid and comprehensive prediction of the temperature field information of a coal-fired boiler under variable operating conditions, being able to monitor and analyze the combustion conditions of the boiler in real time, and timely adjust and optimize the boiler operation mode, so as to realize clean, low-carbon, safe and efficient operation of the coal-fired boiler. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 Flowchart of the method for constructing a digital twin model of the temperature field of a coal-fired boiler provided in this application. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] As digital production, digital operation and digital life gradually become the new normal of society, digital twin technology has been widely studied and applied as an emerging technology that has attracted much attention. In the field of power, power systems tend to be complex and show a trend of big data, which makes digital twin technology gradually penetrate into all aspects of the power industry. Using digital twin technology to build a digital twin model to realize the digital representation of real-world entities or systems, and to realize the digital modeling of the properties, methods, behaviors and other characteristics of physical entities and processes in digital space, it can be used to understand, predict, optimize and control real entities or systems. This enables digital twins to monitor and analyze the operation of boilers in real time and provide timely equipment status feedback. Moreover, digital twin technology can intuitively display the data and models of boilers through visualization, helping engineers to intuitively understand the operation and performance of equipment. Therefore, the present application aims to provide a method for constructing a digital twin model of the temperature field of a coal-fired boiler, which can realize the rapid and comprehensive prediction of the temperature field information of a coal-fired boiler under variable operating conditions, thereby monitoring and analyzing the combustion status of a coal-fired boiler in real time.

[0032] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0033] like Figure 1 As shown, this application takes a 600MW supercritical DC front and rear wall-hedge coal-fired boiler in a power plant as the research object, and proposes a method for constructing a digital twin model of the temperature field of a coal-fired boiler, including:

[0034] Step 1: Obtain the temperature data and input operating parameters of the coal-fired boiler under multi-operating condition numerical simulation.

[0035] Specifically, a numerical combustion model of a coal-fired boiler is established based on the Fluent software, and multiple sub-models such as the radiation heat transfer model, the gas phase turbulent flow model, the coal powder volatilization analysis model, the coke combustion model and the particle motion trajectory model are selected to describe the physical and chemical processes inside the furnace, and the coal quality used is set according to the industrial analysis and elemental analysis of the coal quality in the actual operation of the boiler. The final number of grids is selected based on the grid independence verification results, and the accuracy of the combustion numerical model is verified by comparing the calculation result data of the established combustion numerical model with the actual boiler operation measurement data. As a specific embodiment, the simulation calculation conditions of the combustion numerical model of this application and the actual boiler operation condition parameters are selected from typical condition parameters under a load of 580MW. By comparing the simulation results of the combustion numerical model with the actual operation data, the furnace outlet flue gas temperature, flue gas oxygen content and NO x The relative error between the simulation results of the combustion numerical model and the actual operation data is less than 10%, which meets the accuracy requirements of the simulation.

[0036] Regarding the grid independence verification results and the final number of grids, this application uses the entire furnace as the calculation domain for grid division according to the actual size of the boiler furnace to establish a combustion numerical model. The boiler is divided into 1.91 million, 2.52 million, and 3.13 million grids, respectively. By analyzing the trend of the average flue gas temperature in the furnace with different numbers of grids with height, it is found that the correlation between different grid densities and the temperature values ​​calculated by the combustion numerical model is small, thereby verifying that the simulation results of the combustion numerical model are independent of the number of grids. And because the more grids there are, the lower the calculation efficiency and the higher the calculation cost, this application comprehensively considers the calculation efficiency and calculation accuracy, and finally selects the number of grids as 2.52 million.

[0037] After the verification is completed, under different boiler loads, different actual operating parameters of the boiler (i.e. input operating parameters), such as the amount of coal burned by each layer of the coal-fired boiler burner, the primary air volume of each layer of the burner, the secondary air volume of each layer of the burner, and the burnt air volume of each layer of the burner nozzle, are combined as numerical simulation boundary conditions. Unequal coal and air distribution and equal coal and air distribution operating parameters of power plants in the range of 450MW-580MW are selected, and multi-operating condition numerical simulation is carried out based on the established combustion numerical model to obtain the temperature data of the coal-fired boiler under multi-operating condition numerical simulation.

[0038] Step 2: Use the metric-driven data enhancement method to expand the temperature data and obtain multi-condition temperature field data.

[0039] After obtaining the temperature data of the coal-fired boiler under multi-condition numerical simulation through step 1, the metric-driven data enhancement method is used to expand the data sparse intervals in the temperature data in a targeted manner to complete the collection of multi-condition temperature field data. The principle of the metric-driven data enhancement method is as follows:

[0040] 1) Calculate the mean and standard deviation of the temperature data, and standardize the temperature data so that the temperature data has the same scale. Then calculate the distance between adjacent rows of temperature data, and use Euclidean distance to compare the distance or difference between temperature data in the feature space as the difference measurement result. The Euclidean distance calculation formula is:

[0041]

[0042] Among them, d(x w ,x w+1 ) is the wth temperature data x w and the w+1th temperature data x w+1 The Euclidean distance between them; n is the total dimension of the temperature data; is the wth temperature data x w The v dimension; is the w+1th temperature data x w+1 The v dimension.

[0043] 2) According to the difference measurement result d(x w ,x w+1 ), find out the data sparse interval, so as to achieve the purpose of targeted data expansion and reduce the time cost of data expansion. The search method is as follows:

[0044]

[0045] Among them, δ s is the data sparse interval threshold, used to determine the data sparse interval; W is the total number of temperature data. If the difference measurement result d(x w ,x w+1 ) satisfies formula (2), then the temperature data x w and x w+1 The interval between them is the data sparse interval.

[0046] 3) Perform data expansion on the searched data sparse interval. Since the existing data expansion method may generate invalid redundant data, resulting in a decrease in data quality, it is necessary to determine whether the inserted data is valid.

[0047] Assume that the data sparse interval (temperature data x w and x w+1 The temperature data on the left and right boundaries of the data sparse interval correspond to the boundary conditions of the numerical simulation. By finding the power plant operating parameters within the numerical simulation boundary conditions of the data sparse interval boundary as the numerical simulation boundary conditions, the numerical simulation is performed again to obtain the temperature data (expanded data x nw ), if the extended data x nw If the judgment condition shown in formula (3) is satisfied and the maximum value of the difference measurement result with the boundary of the data sparse interval is smaller than the difference measurement result of the original interval, then the inserted extended data x nw If valid, the data is added to the original temperature data set. Otherwise, the inserted extended data x nw Invalid, cancel data expansion for this interval.

[0048] Specifically, after data expansion is completed, the judgment method is as follows:

[0049] max(d(x w ,x nw ),d(x nw ,x w+1 ))<d(x w ,x w+1 ) (3)

[0050] Among them, d(x w ,x nw ) is xw and x nw The difference measurement results between nw ,x w+1 ) is x nw and x w+1 The difference measurement results between .

[0051] 4) This application can also use information entropy and GINI index to quantitatively evaluate the amount of information in the temperature data after data augmentation and enhancement. By comparing the information entropy and GINI index before and after data augmentation, it can be reflected whether data augmentation can increase the diversity of the data set. If the values ​​of the two indicators after data augmentation are larger than those of the original data, it means that the targeted expansion of the data set can select appropriate samples for expansion according to the characteristics of the data, which increases the diversity of the data and improves the data quality.

[0052] Assuming a given data set X, the information entropy and GINI index are calculated as follows:

[0053]

[0054]

[0055] Where H(X) is the information entropy of data set X; GI(X) is the GINI index of data set X; m a The data x of the ath attribute in the dataset X a , a=1, 2, …, z; h is the total number of data in the data set X.

[0056] Step 3: Based on the temperature field data of multiple working conditions, the modes and modal coefficients are obtained using the eigenorthogonal decomposition method.

[0057] Construct snapshot matrix U based on temperature field data of multiple working conditions L×N , where L is the number of grids in the selected temperature field section, and N is the number of boiler operating conditions, that is, the total number of modes, which can also be directly written as U. The proper orthogonal decomposition method (POD) is used to obtain the modes and modal coefficients. And considering the number of modes and modal energy, high-energy modes are selected to extract and reconstruct the boiler temperature field section. Specifically, the POD principle is as follows:

[0058] 1) Calculate the average value of each node of the snapshot matrix U, that is:

[0059]

[0060] in, is the mean column vector with L rows and 1 column.

[0061] The pulsation matrix of the temperature field is calculated as follows:

[0062]

[0063] in, is the pulsation matrix The i-th column of represents the pulsation vector of each temperature data; i is the i-th column of the snapshot matrix U.

[0064] 2) According to the pulsation matrix Solve the basis function modes and modal coefficients: Calculate the correlation matrix R and find its eigenvalues ​​and eigenvectors. The calculation method is as follows:

[0065]

[0066] Rβ i =λ i β i , (i=1,…,N) (9)

[0067] in, λ i and β i are the eigenvalues ​​and eigenvectors of the correlation matrix R, respectively.

[0068] The POD modes and modal coefficients are calculated as follows:

[0069]

[0070]

[0071] in, is an L×1-dimensional vector, representing the i-th mode; θ i is a 1×N dimensional vector, representing the i-th mode The corresponding modal coefficients.

[0072] 3) With modality The corresponding eigenvalue λ i The size of represents the amount of system energy captured by the basis function mode. Therefore, the contribution of the basis function mode to the total energy of the system is expressed numerically through the energy contribution rate and the cumulative energy contribution rate. The calculation method is as follows:

[0073]

[0074]

[0075] Among them, M< <N,η i Represents the i-th mode Energy contribution rate; E MIt represents the cumulative energy contribution rate of the first M modes. At this time, the first M modes are the main modes.

[0076] The POD modes are arranged in descending order according to their energy ratios, and the first M main modes are selected to describe the entire temperature field data set to achieve the purpose of dimensionality reduction while ensuring reconstruction accuracy. The reconstruction method is:

[0077]

[0078] in, Represents the temperature field snapshot matrix after POD reconstruction.

[0079] Step 4: Based on the input operating parameters and modal coefficients, the prediction model is constructed using the improved online least squares support vector machine algorithm.

[0080] A fitting model (prediction model) is established between the input operating parameters in step 1 and the modal coefficients corresponding to each order mode to achieve rapid prediction of the boiler temperature field. The prediction model is established using an improved online least squares support vector machine, and three update strategies, namely, adding, replacing, and deleting, are introduced to save the existing training parameters. When the prediction model deviation exceeds the set value, the prediction model parameters are updated and corrected by selecting an update strategy to meet the set accuracy again and adapt to changes in nonlinear object characteristics.

[0081] The data set of the prediction model consists of input operating parameters and modal coefficients. Therefore, the 23-dimensional input operating parameters are used as the input of the prediction model, and the 1-dimensional modal coefficient is used as the output of the prediction model. The principle of the improved online least squares support vector machine is as follows:

[0082] 1) Assume that the existing data set T = {(x 1 ,y 1 ),…,(x l ,y l )},in represents the kth input operating condition parameter, and l is the total number of input operating condition parameters; Represents the modal coefficient corresponding to the kth input operating condition parameter. The decision function is constructed as follows:

[0083]

[0084] Among them, α k is the Lagrange multiplier corresponding to the kth input operating condition parameter; K(x,x k ) is the kernel function; b is the deviation.

[0085] Calculate the Lagrange multiplier and deviation according to formula (16):

[0086]

[0087] Where y=[y 1 ,y 2 ,…,y l ] T ; α=[α 1 ,α 2 ,…,α l ] T ; The positive definite matrix H is:

[0088]

[0089] Among them, c is the regularization parameter.

[0090] 2) Calculate the newly added input operating condition parameter x new The Euclidean distance between each input operating condition parameter in the original input operating condition parameter is used to find the input operating condition parameter with the closest Euclidean distance. The distance between the two is d. When the prediction model fails, that is, when the deviation between the predicted output of the prediction model and the actual output is large, use the new input operating condition parameter x new The input operating condition parameter closest to it is replaced or directly added to the input operating condition parameter set, and the parameters of the corresponding prediction model, namely the Lagrange multiplier and the bias, are updated so that the prediction model can adapt to the new input.

[0091] When the prediction model fails, if d <d s (d s The replacement strategy is executed if the distance value is set and the standard for determining whether to execute the replacement strategy or the new strategy is used. The row and column corresponding to the kth input operating condition parameter in the positive definite matrix H are exchanged with the last row and the last column in H to obtain H 1 , as shown in formula (18).

[0092] in, and Respectively, they represent swapping the kth and lth rows and the kth and lth columns of the identity matrix I.

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] Finally, the updated Lagrange multiplier α is calculated by formula (27): * and deviation b * :

[0103]

[0104] Where y*=[y 1 ,…,y k-1 ,y l ,y k+1 ,…,y l-1 ,y new ] T ;

[0105]

[0106] 3) When the model fails, if d>d s , then execute the new strategy. new ,y new ) is added to the original sample set and calculated And update the Lagrange multiplier and deviation of the prediction model to get the new and b * , the update method is shown in formula (28) and formula (29):

[0107]

[0108]

[0109] Among them, at this time y*=[y 1 ,y 2 ,…,y l ,y new ] T .

[0110] 4) Due to the adoption of the new addition strategy, as the Lagrange multipliers and deviations of the prediction model are updated, huge modeling samples will be generated, resulting in an increase in the amount of calculation of the prediction model, affecting the online update speed of the prediction model. Therefore, this application can reduce the consumption of data storage space and improve the calculation speed of the prediction model by adopting the deletion strategy. When the number of samples exceeds the set maximum number of samples N s When , the strategy is executed. After the new sample is added to the sample set, the parameters of formula (28) are calculated to find the two most similar samples in the historical data, and the old samples are deleted in chronological order. Assume that the qth sample needs to be deleted, and the matrix The row and column corresponding to the qth sample in The last row and the last column are swapped to get As shown in formula (30).

[0111]

[0112] Use formula (31) to calculate the updated Lagrange multiplier α * and deviation b * :

[0113]

[0114] in, The calculation of is based on formula (21).

[0115] Step 5: Use the prediction model to predict the current modal coefficients of each order corresponding to the current input operating condition parameters.

[0116] After the prediction model is constructed in step 4 above, the current input operating condition parameters are input into the prediction model to predict the current modal coefficients θ' of each order corresponding to the current input operating condition parameters.

[0117] Step 6: Construct a digital twin model of the temperature field of the coal-fired boiler based on the modes and current modal coefficients of each order.

[0118] According to the prediction model constructed by the improved online least squares support vector machine in step 5, the current modal coefficients θ' of each order corresponding to the current input operating parameters are predicted, and finally the digital twin model of the temperature field of the coal-fired boiler can be constructed using formula (14). Specifically, the predicted current modal coefficients θ' of each order are multiplied by the main mode determined in step 3 using formula (14) and then added with the average value of each node of the snapshot matrix to obtain the reconstructed temperature field snapshot matrix, which can accurately map the boiler temperature field and realize the prediction of the temperature field. The formula of the reconstructed temperature field snapshot matrix U" is as follows:

[0119]

[0120] in, is the sth main mode. A temperature field digital twin model is constructed based on the reconstructed temperature field snapshot matrix U". Specifically, the temperature data generated by CFD simulation is combined with the reconstructed temperature field snapshot matrix obtained by using intrinsic orthogonal decomposition and an improved online least squares support vector machine prediction model to form the temperature field digital twin model of this application. The temperature field digital twin model of this application is used to accurately predict the temperature field of a coal-fired boiler.

[0121] In addition, this application also uses the mean absolute percentage error MAPE, root mean square error RMSE and determination coefficient R2 The accuracy of the temperature field digital twin model is comprehensively quantified and the fitting ability of the temperature field digital twin model is evaluated to ensure that the temperature field digital twin model of this application has high speed and accuracy.

[0122] In summary, this application proposes a method for constructing a digital twin model of the temperature field of a coal-fired boiler. In order to make the data samples cover multiple working conditions more comprehensively, a metric-driven data enhancement method is adopted to improve the quality of sample data and reduce data redundancy while minimizing the number of CFD simulations. At the same time, in order to reduce the computational time required to obtain temperature field information, under the premise of ensuring the calculation accuracy, the reduced-order model method is used to replace the complex full-order CFD model of the temperature field system with a reduced-order model that is much smaller than the order of the original temperature field system. A digital twin model of the boiler temperature field is constructed based on CFD data and the reduced-order method in combination with a machine learning algorithm to achieve rapid prediction of the boiler temperature field information under variable working conditions, thereby real-time monitoring and analysis of the boiler's combustion conditions, timely adjustment and optimization of the boiler operation mode, and clean, low-carbon, safe and efficient operation of coal-fired boilers.

[0123] Compared with the prior art, the method for constructing a digital twin model of the temperature field of a coal-fired boiler in this application has the following advantages:

[0124] (1) Based on Fluent fluid dynamics software and combined with the metric-driven data enhancement method, the actual boiler combustion process was numerically simulated to obtain boiler temperature distribution data under various working conditions. This numerical simulation method can overcome the limitations of traditional measurement methods such as the on-site environment, complex equipment structure, and high installation and maintenance costs. It has the advantages of mature technology, low cost, and complete data. At the same time, the use of metric-driven data enhancement methods can effectively reduce the number of CFD simulations, improve sample data quality, and reduce data redundancy.

[0125] (2) By extracting the main characteristic modes using POD, the temperature field can be reconstructed through a small number of modes, achieving order reduction while better describing the temperature distribution. The original high-dimensional complex high-order system is replaced by a low-dimensional system characterized by the principal component mode fitting, greatly reducing the computational cost.

[0126] (3) The digital twin model of the temperature field of a coal-fired boiler that combines POD and an improved online least squares support vector machine can reliably predict the boiler temperature field, effectively solve the problem of high computational time cost of Fluent, and quickly and accurately map the internal temperature distribution information of the boiler while visualizing it. This has important guiding significance for the optimization of power plant operation, thereby promoting the development of boiler combustion monitoring towards high-end and intelligent directions, and realizing the integration of the digital revolution and the energy revolution.

[0127] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0128] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for constructing a digital twin model of a coal-fired boiler temperature field, characterized in that: include: Acquire temperature data and input operating condition parameters of a coal-fired boiler under multi-operating condition numerical simulation; the input operating condition parameters include the amount of coal burned by burners on each layer of the coal-fired boiler, the primary air volume of burners on each layer, the secondary air volume of burners on each layer, and the burnt air volume of burnt air nozzles on each layer; The temperature data is expanded using a metric-driven data enhancement method to obtain temperature field data under multiple working conditions. Based on the temperature field data of multiple working conditions, the modes and modal coefficients are obtained using the eigenorthogonal decomposition method; Based on the input operating parameters and modal coefficients, the prediction model is constructed using an improved online least squares support vector machine algorithm; The current modal coefficients of each order corresponding to the current input working condition parameters are predicted by the prediction model; A temperature field digital twin model of a coal-fired boiler is constructed based on modes and current modal coefficients of each order; the temperature field digital twin model is used to predict the temperature field.

2. The method for constructing a digital twin model of a coal-fired boiler temperature field according to claim 1, characterized in that: The metric-driven data enhancement method is used to expand the temperature data to obtain multi-condition temperature field data, specifically including: Based on the wth temperature data x w and the w+1th temperature data x w+1 , using the formula Calculate x w and x w+1 The difference measurement result d(x w ,x w+1 ), where n is the total dimension of the temperature data; is the wth temperature data x w The v dimension; is the w+1th temperature data x w+1 The v dimension; According to the difference measurement result d(x w ,x w+1 ), through the formula Determine the data sparse interval and perform data expansion; where W is the total number of temperature data; δ s is the data sparse interval threshold.

3. The method for constructing a digital twin model of a coal-fired boiler temperature field according to claim 2, characterized in that: After the temperature data is expanded using the metric-driven data enhancement method, the method further includes: After data expansion is completed, if the expanded data x nw Satisfies the formula max(d(xw,xnw),d(xnw,xw +1 ))<d(xw,xw +1 ), then the expanded data xnw is valid; otherwise, the expanded data x nw Invalid, cancel data expansion; where d(x w ,x nw ) is x w and x nw The difference measurement results between nw ,x w+1 ) is x nw and x w+1 The difference measurement results between .

4. The method for constructing a digital twin model of a coal-fired boiler temperature field according to claim 1, characterized in that: The method of obtaining the modes and modal coefficients based on the temperature field data of multiple working conditions by using the intrinsic orthogonal decomposition method specifically includes: Based on the temperature field data of multiple working conditions, the intrinsic orthogonal decomposition method is used through the formula Calculate the i-th mode By formula Calculate the i-th mode The corresponding modal coefficient θ i ; Where N is the total number of modes; is the pulsation matrix; i and β i are the eigenvalues ​​and eigenvectors of the correlation matrix R, 5. The method for constructing a digital twin model of a coal-fired boiler temperature field according to claim 1, characterized in that: The prediction model is constructed based on the input operating condition parameters and modal coefficients by using an improved online least squares support vector machine algorithm, which specifically includes: Based on the input operating parameters and modal coefficients, with 23-dimensional input operating parameters as the prediction model input and 1-dimensional modal coefficient as the prediction model output, the improved online least squares support vector machine algorithm is used to build and train the prediction model.

6. The method for constructing a digital twin model of a coal-fired boiler temperature field according to claim 5, characterized in that: The decision function of the improved online least squares support vector machine algorithm is Among them, α k is the Lagrange multiplier corresponding to the kth input operating condition parameter; K(x,x k ) is the kernel function; b is the deviation; l is the total number of input operating condition parameters.

7. The method for constructing a digital twin model of a coal-fired boiler temperature field according to claim 1, characterized in that: The digital twin model of the temperature field of the coal-fired boiler is constructed based on the modal and the current modal coefficients of each order, specifically including: Based on the mode, the number of main modes is determined using the formula Get the reconstructed temperature field snapshot matrix U"; where, is the sth main mode; θ' is the current modal coefficient of each order; M is the total number of main modes; is the mean column vector; A digital twin model of the temperature field of a coal-fired boiler is constructed based on the reconstructed temperature field snapshot matrix.

Images