Method, system and equipment for acquiring internal operation data of boiler furnace

Abstract

The invention discloses a method, a system and equipment for acquiring internal operation data of a boiler hearth, belongs to the technical field of data processing, and solves the problems of low accuracy and lack of interpretation of an existing method for acquiring the internal operation data of the hearth. The method comprises the following steps: embedding prior knowledge and mechanism satisfied in a boiler operation process in a machine learning process for controlling and correcting the machine learning process, and improving the accuracy of a predicted boiler operation data result, that is, a trained result accords with a physical rule. A mass conservation mechanism is embedded in a stable flue gas component area of a boiler tail heating surface, so that the interpretability and rationality of a machine learning training result are improved, and machine learning prediction data are more in line with the actual operation rule of the boiler and fit the reality. The method is suitable for acquiring the internal operation data of the hearth during boiler operation.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to the acquisition of operating data inside a boiler furnace. Background Art

[0002] With the rapid development of digital information technology, machine learning has been widely used in various fields because of its characteristics of large information processing volume, high data processing efficiency, and high data calculation accuracy. During the operation of the boiler, due to its complex operating state, the operator cannot obtain the operating data inside the furnace in a timely and accurate manner. Therefore, it is necessary to obtain the operating data through machine learning. Under stable operating conditions, machine learning can obtain the operating data inside the boiler furnace in a timely and accurate manner. However, due to its operating characteristics, the boiler generates a large amount of data, various types of data, and a wide range of data changes. Traditional machine learning driven by massive data is difficult to achieve good results. It often happens that although such a result is obtained, it is impossible to analyze in more depth why this result can be obtained. Therefore, when the boiler operating load fluctuates greatly, when poor learning results are obtained through traditional machine learning methods, the reasons for the poor learning results cannot be explained, and thus it is impossible to further take effective measures to solve the problem of poor learning results. Summary of the invention

[0003] The purpose of the present invention is to solve the problems of low accuracy and lack of interpretability of the existing method for obtaining the internal operation data of the furnace, and to provide a method, system and device for obtaining the internal operation data of the boiler furnace.

[0004] The present invention is implemented by the following technical solutions. On the one hand, the present invention provides a method for obtaining internal operation data of a boiler furnace, the method comprising:

[0005] Step 1: Obtain the operating parameters of the loads of several working conditions;

[0006] Step 2: Select two cross sections near the economizer outlet, with n nodes in each cross section;

[0007] Step 3: Construct training data for machine learning, and form an input parameter matrix with input coordinates and operating parameters;

[0008] Step 4: Normalize the training data;

[0009] Step 5: Use machine learning to train and build a prediction model;

[0010] Step 6: Check whether the machine learning results satisfy the law of conservation of carbon between the cross-sections and nodes, and correct the machine learning results using the law of conservation of carbon, including:

[0011] Step 6.1: Take the economizer and its vicinity as the control volume, and take the inlet and outlet of the area as the cross-sections that satisfy the carbon conservation law, that is, the carbon mass at the inlet and outlet nodes is equal, and the mass of carbon entering the furnace is equal to the mass of carbon in each product;

[0012] Step 6.2: Perform cross-sectional carbon conservation test to eliminate the face-to-face errors of the three carbon-containing substances;

[0013] Step 6.3: Conduct nodal carbon conservation checks to eliminate point-to-point deviations of the three carbon-containing substances;

[0014] Step 7: Use the prediction model corrected in step 6 to obtain the internal operation data of the boiler furnace.

[0015] Furthermore, the operating parameters in step 1 include: load, total air volume, total coal volume, primary air volume, secondary air volume, secondary air temperature, coal mill operation mode, coal mill swing angle, outlet flue gas mass flow rate, unburned carbon C of each grid node, B Mass flow rate, CO and CO2 mass fraction.

[0016] Further, step 3 includes:

[0017] The input coordinates and operating parameters form an input parameter matrix (x, y, z, fuhe, TF, TA, T1A, T2A,, W, M, J), where x, y, z are the three-dimensional coordinate values, fuhe is the load, TF is the total coal volume, TA is the total air volume, T1A is the total primary air volume, T2A is the total secondary air volume, W is the temperature, M is the height of the coal mill, and J is the burnout air swing angle;

[0018] The output coordinates and nitric oxide ppm, temperature, and carbon-containing component mass fraction form the output parameter matrix (x, y, z, m CB ,1e4*m CO ,m CO2 , T, NO), where T is temperature, NO is nitric oxide concentration, m CB is the unburned carbon, 1e4*m CO is carbon monoxide, m CO2 For carbon dioxide.

[0019] Further, step 5 comprises:

[0020] Step 5.1: Initialize the ELM model;

[0021] Step 5.2: Add neurons. W and B are automatically randomly generated by default.

[0022] Step 5.3: Train the ELM model using the labeled training set;

[0023] Step 5.4: Input the test data and give the predicted value on the test set;

[0024] Step 5.5: Use mean square error (MSE) and relative error (EPS) to evaluate the accuracy of the model.

[0025] Further, step 6.1 includes:

[0026]

[0027] Among them, Y CO , is the gas mass fraction, is the carbon mass flow rate at point (x,y,z), is the mass flow rate of unburned carbon, is the mass flow rate of carbon monoxide, is the mass flow rate of carbon dioxide; is the flue gas mass flow rate;

[0028] The carbon conservation equation is:

[0029]

[0030] in, is the divergence of carbon mass flow, m' C " is the carbon source term, which is 0 for the economizer region. The equations applied to this region include:

[0031] Cross-sectional carbon conservation equation:

[0032] Nodal carbon conservation equation:

[0033] Further, step 6.2 includes:

[0034] Step 6.2.1: Carry out cross-section carbon conservation test, the test criteria are: b T T test a=0

[0035] in,

[0036]

[0037] Step 6.2.2: Calculate the face-to-face deviation of the three carbon-containing substances, including the unburned carbon m CB , carbon monoxide 1e4*m CO and carbon dioxide m CO2 , specifically including:

[0038] Use the ELM model to predict the output variable T, T_test = Model.predict(X_test);

[0039] Find the deviation of the three carbon-containing substances from each other:

[0040] e1 = numpy.dot(bT,m CB ), numpy is a database calling algorithm in python, dot is a function in the algorithm, which is used to sum the unburned carbon in each surface and then make the difference between the two surfaces;

[0041] e2 = numpy.dot(bT,m CO ), which is used to sum the carbon monoxide in each face and then make the difference between the two faces;

[0042] e3 = numpy.dot(bT,m CO2 ), which is used to sum the carbon dioxide in each surface and then make the difference between the two surfaces;

[0043] Step 6.2.3: Assume that the ELM model is accurate in predicting the average value of each substance at 2n points, so the two surfaces each bear half of the correction; this half of the correction is evenly distributed among the n points in the surface: T1 is the output variable matrix after correction, T is the output variable matrix before correction, and the deviation correction algorithm for the three substances is as follows:

[0044] Export: T1[:,0]=T[:,0]-1 / 2*e1 / n

[0045] Entry: T1[:,0]=T[:,0]+1 / 2*e1 / n

[0046] Export: T1[:,1]=T[:,1]-1 / 2*e2 / n

[0047] Entry: T1[:,1]=T[:,1]+1 / 2*e2 / n

[0048] Export: T1[:,2]=T[:,2]-1 / 2*e3 / n

[0049] Entry: T1[:,2]=T[:,2]+1 / 2*e3 / n.

[0050] Further, step 6.3 includes:

[0051] Step 6.3.1: Conduct node carbon conservation test, the test criteria are: in P=I 6×6 -aa(aa T aa) -1 , T yis the node temperature value on the y-axis section, T y+dy is the node temperature value on the section with ordinate y+dy;

[0052] Step 6.3.2: Calculate the point-to-point deviations of the three carbon-containing substances, including:

[0053] Use the ELM model to predict the output variable T, T_test = Model.predict(X_test);

[0054] Construct matrix bb for point-by-point subtraction:

[0055]

[0056] Find the point-to-point deviations of the three carbon-containing species:

[0057] e4=abs(np.dot(bb,T_test[:,0])).max(axis=0), find the n positions with the largest absolute deviation of unburned carbon.

[0058] e5 = abs(np.dot(bb,T_test[:,1])).max(axis = 0), find the n positions with the largest absolute deviation of carbon monoxide,

[0059] e6 = abs(np.dot(bb,T_test[:,2])).max(axis = 0), find the n positions with the largest absolute deviation of carbon dioxide;

[0060] Find the maximum absolute value of the total carbon deviation at n positions:

[0061] abs(np.dot(np.dot(bb,nT_test),[112 / 2812 / 44])).max(axis=0);

[0062] Step 6.3.3: Eliminate the deviations found in the above steps through the conservation of carbon:

[0063] Use the ELM model to predict the output T, T_test = Model.predict(X_test), where T_test is the output value of the prediction parameter, and X_test is the corresponding input parameter of the prediction parameter;

[0064] Write T as an n*6-dimensional matrix TT:

[0065]

[0066] Multiply the projection matrix P so that aa is in In the null space of:

[0067]

[0068] Write back as a 2n*3 dimensional matrix The matrix of node temperature values ​​after eliminating deviation.

[0069] In a second aspect, the present invention provides a system for acquiring internal operation data of a boiler furnace, the system comprising:

[0070] An operating parameter acquisition module is used to obtain operating parameters of loads under several working conditions;

[0071] The section selection module is used to select two sections near the economizer outlet, with n nodes in each section;

[0072] The learning data construction module is used to construct the training data for machine learning, and to form an input parameter matrix with input coordinates and operating parameters;

[0073] Normalization module, used to normalize the training data;

[0074] Prediction model building module, used to use machine learning for learning and training to build a prediction model;

[0075] The prediction model correction module is used to check whether the machine learning results satisfy the law of conservation of carbon between the cross-section and the node, and to correct the machine learning results by the law of conservation of carbon, including:

[0076] Step 6.1: Take the economizer and its vicinity as the control volume, and take the inlet and outlet of the area as the cross-sections that satisfy the carbon conservation law, that is, the carbon mass at the inlet and outlet nodes is equal, and the mass of carbon entering the furnace is equal to the mass of carbon in each product;

[0077] Step 6.2: Perform cross-sectional carbon conservation test to eliminate the face-to-face errors of the three carbon-containing substances;

[0078] Step 6.3: Conduct nodal carbon conservation checks to eliminate point-to-point deviations of the three carbon-containing substances;

[0079] The operation data acquisition module is used to obtain the internal operation data of the boiler furnace using the revised prediction model.

[0080] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the steps of a method for obtaining operating data inside a boiler furnace as described above are executed.

[0081] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein a plurality of computer instructions are stored in the computer-readable storage medium, and the plurality of computer instructions are used to enable a computer to execute a method for acquiring operating data inside a boiler furnace as described above.

[0082] Beneficial effects of the present invention:

[0083] The present invention provides a method for acquiring the internal operation data of a boiler furnace by embedding a boiler mechanism in a machine learning process, thereby increasing the interpretability of the boiler operation data predicted by machine learning, and the rationality and accuracy of the results. Prior knowledge and mechanisms satisfied in the boiler operation process are embedded in the machine learning process to control and correct the machine learning process, and to improve the accuracy of the predicted boiler operation data results, that is, the training results are in line with the laws of physics. By embedding the mass conservation mechanism in the stable region of the flue gas composition on the rear heating surface of the boiler, the interpretability and rationality of the machine learning training results are increased, making the machine learning prediction data more in line with the actual operation laws of the boiler and more in line with reality.

[0084] The present invention is suitable for obtaining the internal operation data of the furnace when the boiler is running. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0086] Figure 1 A schematic diagram of the process of embedding boiler mechanism into machine learning results according to the method of the present invention;

[0087] Figure 2 is the temperature calculation result;

[0088] Figure 3 is the result of calculation of nitrogen oxides;

[0089] Figure 4 This is the machine learning machine structure embedded with the boiler mechanism of the present invention. DETAILED DESCRIPTION

[0090] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0091] Embodiment 1: A method for acquiring internal operation data of a boiler furnace, the method comprising:

[0092] Step 1: Obtain the operating parameters of the loads of several working conditions;

[0093] Step 2: Select two cross sections near the economizer outlet, with n nodes in each cross section;

[0094] Step 3: Construct training data for machine learning, and form an input parameter matrix with input coordinates and operating parameters;

[0095] Step 4: Normalize the training data;

[0096] Step 5: Use machine learning to train and build a prediction model;

[0097] Step 6: Check whether the machine learning results satisfy the law of conservation of carbon between the cross-sections and nodes, and correct the machine learning results using the law of conservation of carbon, including:

[0098] Step 6.1: Take the economizer and its vicinity as the control volume, and take the inlet and outlet of the area as the cross-sections that satisfy the carbon conservation law, that is, the carbon mass at the inlet and outlet nodes is equal, and the mass of carbon entering the furnace is equal to the mass of carbon in each product;

[0099] Step 6.2: Perform cross-sectional carbon conservation test to eliminate the face-to-face errors of the three carbon-containing substances;

[0100] Step 6.3: Conduct nodal carbon conservation checks to eliminate point-to-point deviations of the three carbon-containing substances;

[0101] Step 7: Use the prediction model corrected in step 6 to obtain the internal operation data of the boiler furnace.

[0102] This implementation method proposes a method for embedding boiler mechanisms in the machine learning process. By adding boiler mechanisms to the traditional machine learning process, the error uncertainty of traditional big data-driven machine learning predictions of on-site operating data and the failure to meet boiler operating laws are eliminated, and the accuracy of traditional machine learning prediction results is further improved to meet boiler operating laws, thereby providing effective operating suggestions for optimizing the boiler operating status.

[0103] Embodiment 2: This embodiment further limits the method for obtaining the internal operation data of the boiler furnace as described above. In this embodiment, the operation parameters in step 1 are further limited, specifically including:

[0104] The operating parameters in step 1 include: load, total air volume, total coal volume, primary air volume, secondary air volume, secondary air temperature, coal mill operation mode, coal mill swing angle, outlet flue gas mass flow rate, unburned carbon C of each grid node, BMass flow rate, CO and CO2 mass fraction.

[0105] Implementation method 3: This implementation method further limits the method for obtaining the internal operation data of the boiler furnace as described above. In this implementation method, step 3 is further limited, specifically including:

[0106] Step 3 specifically includes:

[0107] The input coordinates and operating parameters form an input parameter matrix (x, y, z, fuhe, TF, TA, T1A, T2A,, W, M, J), where x, y, z are the three-dimensional coordinate values, fuhe is the load, TF is the total coal volume, TA is the total air volume, T1A is the total primary air volume, T2A is the total secondary air volume, W is the temperature, M is the height of the coal mill, and J is the burnout air swing angle;

[0108] The output coordinates and nitric oxide ppm, temperature, and carbon-containing component mass fraction form the output parameter matrix (x, y, z, m CB ,1e4*m CO ,m CO2 , T, NO), where T is temperature, NO is nitric oxide concentration, m CB is the unburned carbon, 1e4*m CO is carbon monoxide, m CO2 For carbon dioxide.

[0109] Embodiment 4: This embodiment further limits the method for obtaining the internal operation data of the boiler furnace as described above. In this embodiment, step 5 is further limited, specifically including:

[0110] Step 5 specifically includes:

[0111] Step 5.1: Initialize the ELM model;

[0112] Step 5.2: Add neurons. W and B are automatically randomly generated by default.

[0113] Step 5.3: Train the ELM model using the labeled training set;

[0114] Step 5.4: Input the test data and give the predicted value on the test set;

[0115] Step 5.5: Use mean square error (MSE) and relative error (EPS) to evaluate the accuracy of the model.

[0116] The ELM model of this embodiment can improve the accuracy of the internal operation data of the boiler furnace.

[0117] Embodiment 5: This embodiment further defines the method for obtaining the internal operation data of a boiler furnace as described above. In this embodiment, step 6.1 is further defined, specifically including:

[0118] Step 6.1 includes:

[0119]

[0120] Among them, Y CO , is the gas mass fraction, is the carbon mass flow rate at point (x,y,z), is the mass flow rate of unburned carbon, is the mass flow rate of carbon monoxide, is the mass flow rate of carbon dioxide; is the flue gas mass flow rate;

[0121] The carbon conservation equation is:

[0122]

[0123] in, is the divergence of carbon mass flow, m' C " is the carbon source term, which is 0 for the economizer region. The equations applied to this region include:

[0124] Cross-sectional carbon conservation equation:

[0125] Nodal carbon conservation equation:

[0126] This embodiment provides a specific solution in which the economizer and its vicinity are used as the control body, and the inlet and outlet of the area are respectively taken as the cross-sections that satisfy the conservation of carbon element, so as to ensure that the sum of the C element mass at the inlet and outlet nodes is equal, and the mass of the carbon element entering the furnace is equal to the mass of the carbon element in each product.

[0127] Embodiment 6: This embodiment further defines the method for obtaining the internal operation data of a boiler furnace as described above. In this embodiment, step 6.2 is further defined, specifically including:

[0128] Step 6.2 includes:

[0129] Step 6.2.1: Carry out cross-section carbon conservation test, the test criteria are: b T T test a=0

[0130] in,

[0131]

[0132] Step 6.2.2: Calculate the face-to-face deviation of the three carbon-containing substances, including the unburned carbon m CB , carbon monoxide 1e4*m CO and carbon dioxide m CO2 , specifically including:

[0133] Use the ELM model to predict the output variable T, T_test = Model.predict(X_test);

[0134] Find the deviation of the three carbon-containing substances from each other:

[0135] e1 = numpy.dot(bT,m CB ), numpy is a database calling algorithm in python, dot is a function in the algorithm, which is used to sum the unburned carbon in each surface and then make the difference between the two surfaces;

[0136] e2 = numpy.dot(bT,m CO ), which is used to sum the carbon monoxide in each face and then make the difference between the two faces;

[0137] e3 = numpy.dot(bT,m CO2 ), which is used to sum the carbon dioxide in each surface and then make the difference between the two surfaces;

[0138] It should be noted that the deviation of the cross-sectional carbon conservation equation is calculated as follows: e = numpy.dot(numpy.dot(bT, T_test), a), the total carbon of the two surfaces is subtracted, and the three substances used in the test each correct their own deviation e i , which is equivalent to the conservation of each substance. In fact, it only requires that the total correction is e;

[0139] Step 6.2.3: Assume that the ELM model is accurate in predicting the average value of each substance at 2n points, so the two surfaces each bear half of the correction; this half of the correction is evenly distributed among the n points in the surface: T1 is the output variable matrix after correction, T is the output variable matrix before correction, and the deviation correction algorithm for the three substances is as follows:

[0140] Export: T1[:,0]=T[:,0]-1 / 2*e1 / n

[0141] Entry: T1[:,0]=T[:,0]+1 / 2*e1 / n

[0142] Export: T1[:,1]=T[:,1]-1 / 2*e2 / n

[0143] Entry: T1[:,1]=T[:,1]+1 / 2*e2 / n

[0144] Export: T1[:,2]=T[:,2]-1 / 2*e3 / n

[0145] Entry: T1[:,2]=T[:,2]+1 / 2*e3 / n.

[0146] This embodiment provides a method for eliminating the face-to-face deviation of three carbon-containing substances.

[0147] Embodiment 7: This embodiment further defines the method for obtaining the internal operation data of a boiler furnace as described above. In this embodiment, step 6.3 is further defined, specifically including:

[0148] Step 6.3 includes:

[0149] Step 6.3.1: Conduct node carbon conservation test, the test criteria are: in P=I 6×6 -aa(aa T aa) -1 , T y is the node temperature value on the y-axis section, T y+dy is the node temperature value on the section with ordinate y+dy;

[0150] Step 6.3.2: Calculate the point-to-point deviations of the three carbon-containing substances, including:

[0151] Use the ELM model to predict the output variable T, T_test = Model.predict(X_test);

[0152] Construct matrix bb for point-by-point subtraction:

[0153]

[0154] Find the point-to-point deviations of the three carbon-containing species:

[0155] e4=abs(np.dot(bb,T_test[:,0])).max(axis=0), find the n positions with the largest absolute deviation of unburned carbon.

[0156] e5 = abs(np.dot(bb,T_test[:,1])).max(axis = 0), find the n positions with the largest absolute deviation of carbon monoxide,

[0157] e6 = abs(np.dot(bb,T_test[:,2])).max(axis = 0), find the n positions with the largest absolute deviation of carbon dioxide;

[0158] Find the maximum absolute value of the total carbon deviation at n positions:

[0159] abs(np.dot(np.dot(bb,nT_test),[112 / 2812 / 44])).max(axis=0);

[0160] Step 6.3.3: Eliminate the deviations found in the above steps through the conservation of carbon:

[0161] Use the ELM model to predict the output T, T_test = Model.predict(X_test), where T_test is the output value of the prediction parameter, and X_test is the corresponding input parameter of the prediction parameter;

[0162] Write T as an n*6-dimensional matrix TT:

[0163]

[0164] Multiply the projection matrix P so that aa is in In the null space of:

[0165]

[0166] Write back as a 2n*3 dimensional matrix The matrix of node temperature values ​​after eliminating deviation.

[0167] This embodiment provides a method for eliminating point-to-point deviations of three carbon-containing substances.

[0168] Embodiment 8: This embodiment is a specific example of a method for acquiring internal operation data of a boiler furnace as described above, and specifically includes:

[0169] In order to facilitate the understanding of the method of embedding boiler mechanism in the machine learning process mentioned in the invention patent, the following example is taken as the research object of a 350MW four-corner tangential boiler:

[0170] S1: Known data:

[0171] The operating parameters of the loads of 27 working conditions, including load, total air volume, total coal volume, primary air volume, secondary air volume, secondary air temperature, coal mill operation mode, coal mill swing angle, etc., as well as the outlet flue gas mass flow rate, unburned carbon C of each grid node B Mass flow rate, CO and CO2 mass fraction.

[0172] S2: Simplified analysis: Select two cross-sections near the economizer outlet, with n nodes in each cross-section.

[0173] S3: Build machine learning training data and output data:

[0174] The input coordinates and operating parameters are used to form an input parameter matrix (x, y, z, fuhe, TF, TA, T1A, T2A,, W, M, J), where x, y, z are three-dimensional coordinate values, fuhe is the load, TF is the total coal volume, TA is the total air volume, T1A is the total primary air volume, T2A is the total secondary air volume, W is the temperature, M is the height of the coal mill, and J is the burnout wind swing angle. The output coordinates and nitrogen monoxide ppm, temperature, carbon-containing component mass fraction, 3D (unburned carbon m CB , carbon monoxide 1e4*m CO ,CO2m CO2 ) forms the output parameter matrix (x, y, z, m CB ,1e4*m CO ,m CO2 , T, NO), where x, y, z are the three-dimensional coordinate values, T is the temperature, and NO is the nitric oxide concentration.

[0175] S4: Use z-score to normalize the parameters in the above input and output matrices.

[0176] S5: Use machine learning for training and build a prediction model (you can choose the ELM model). The specific process is as follows:

[0177] 1) Initialize an ELM (input, output) model:

[0178] Model = hpelm.ELM(11,3)

[0179] 2) Add neurons, and automatically generate W and B randomly by default:

[0180] Model.add_neurons(50,'sigm')

[0181] 3) Train the ELM model using the labeled training set:

[0182] model.train(X_trian,T_train)

[0183] 4) Input the test data and give the predicted value on the test set:

[0184] Y_test=model.predict(X_test)

[0185] 5) Use mean square error (MSE) and relative error (EPS) to evaluate model accuracy:

[0186] TEST(Y_test,T_test)

[0187] EPS = max(abs((YT) / T))

[0188] 6) Embed the boiler mechanism, and verify whether the machine learning results satisfy the C element conservation law between the cross section and the node, and further correct the machine learning effect through the C element conservation law. The specific process is as follows:

[0189] 6.1) Taking the economizer and its vicinity as the control body, the inlet and outlet of the area are taken as the cross-sections that satisfy the conservation of carbon element, that is, the mass of the C element at the inlet and outlet nodes should be equal, and the mass of the carbon element entering the furnace is equal to the mass of the carbon element in each product:

[0190]

[0191] Among them, Y CO , is the gas mass fraction, is the carbon mass flow rate at point (x, y, z);

[0192] is the mass flow rate of unburned carbon, is the mass flow rate of carbon monoxide, is the mass flow rate of carbon dioxide, and the carbon conservation equation is: is the divergence of carbon mass flow,

[0193] m' C " is the carbon source term, which is 0 for the economizer area. is the flue gas mass flow rate.

[0194] Applied to this region, the equation can be expressed as:

[0195] Cross-sectional carbon conservation equation

[0196] Nodal Carbon Conservation Equation

[0197] 6.2) First, the cross-section carbon conservation test is carried out, and the test criteria are: b T T test a=0

[0198] in,

[0199]

[0200] Next, the calculation of deviation elimination is performed:

[0201] 1. Use the ELM model to predict the output variable T, T_test = Model.predict(X_test). The purpose is to build a machine learning algorithm framework, which is the basis for the implementation of subsequent algorithms.

[0202] 2. Find the face-to-face deviation of the three carbon-containing substances

[0203] e1 = numpy.dot(bT,m CB ), numpy is a database calling algorithm in python, dot is a function in the algorithm, which is used to sum the unburned carbon in each surface and then make the difference between the two surfaces;

[0204] e2 = numpy.dot(bT,m CO ), which is used to sum the carbon monoxide in each face and then make the difference between the two faces;

[0205] e3 = numpy.dot(bT,m CO2 ), which is used to sum the carbon dioxide in each surface and then make the difference between the two surfaces;

[0206] 3. Find the deviation of the cross-sectional carbon conservation equation:

[0207] e=numpy.dot(numpy.dot(bT,T_test),a), the difference between the total carbon of the two surfaces, the three substances used in the test are each corrected for their own deviation e i , which is equivalent to the conservation of each substance. In fact, it only requires that the total correction is e;

[0208] Assuming that ELM's prediction of the average value of each substance at 2n points is accurate, the two surfaces each bear half of the correction; this half of the correction is evenly distributed among the n points in the surface: T1 is the output variable matrix after correction, and T is the output variable matrix before correction. The deviation correction algorithm for the three substances is as follows:

[0209] Export: T1[:,0]=T[:,0]-1 / 2*e1 / n

[0210] Entry: T1[:,0]=T[:,0]+1 / 2*e1 / n

[0211] Export: T1[:,1]=T[:,1]-1 / 2*e2 / n

[0212] Entry: T1[:,1]=T[:,1]+1 / 2*e2 / n

[0213] Export: T1[:,2]=T[:,2]-1 / 2*e3 / n

[0214] Entry: T1[:,2]=T[:,2]+1 / 2*e3 / n

[0215] 6.3) Next, the node carbon conservation test is carried out, and the test criteria are: (a matrix whose elements are all 0)

[0216] in P=I 6×6 -aa(aa T aa) -1 , then the deviation elimination calculation is performed, T y is the node temperature value on the y-axis section, T y+dy It is the node temperature value on the section with vertical coordinate y+dy.

[0217] 1. Use the ELM model to predict the output variable T, T_test = Model.predict(X_test);

[0218] 2. Construct matrix bb for point-by-point difference:

[0219]

[0220] 3. Find the point-to-point deviations of the three carbon-containing substances:

[0221] e4 = abs(np.dot(bb,T_test[:,0])).max(axis = 0), find the n positions with the largest absolute deviation of unburned carbon;

[0222] e5 = abs(np.dot(bb,T_test[:,1])).max(axis = 0), find the n positions with the largest absolute deviation of carbon monoxide;

[0223] e6 = abs(np.dot(bb,T_test[:,2])).max(axis = 0), find the n positions with the largest absolute deviation of carbon dioxide;

[0224] 4. Find the maximum absolute value of the total carbon deviation at n positions, according to the formula Calculate the total carbon deviation, because after the deviations of the three substances are corrected, the total carbon deviation must be corrected:

[0225] abs(np.dot(np.dot(bb,nT_test),[112 / 2812 / 44])).max(axis=0)

[0226] 5. Eliminate all deviations detected in the above steps through the conservation of carbon:

[0227] 5.1. Use the ELM model to predict the output T:

[0228] T_test=Model.predict(X_test), where T_test is the output value of the prediction parameter, and X_test is the corresponding input parameter of the prediction parameter.

[0229] 5.2. Write T as an n*6-dimensional matrix TT:

[0230]

[0231] 5.3. Multiply the projection matrix P so that aa is In the null space of :

[0232]

[0233] 5.4. Write back to 2n*3 dimensional matrix The matrix of node temperature values ​​after eliminating deviation.

[0234] By embedding the boiler machine into the machine learning algorithm learning machine, the typical operating parameters of the boiler, including temperature and nitrogen oxide concentration, are calculated and predicted to verify the generalization of the machine learning algorithm learning machine embedded in the boiler mechanism. The calculation results are as follows Figure 2 and Figure 3 shown.

[0235] Embedding the boiler mechanism into the machine learning algorithm and predicting the typical operating parameters of the boiler can improve the processing capabilities of operating data, enhance the interpretability and accuracy of predicted operating data, and facilitate on-site operating personnel to judge the boiler operating conditions more timely and accurately, thus providing more reliable guarantees for maintaining stable boiler operation and further controlling and optimizing boiler operation.

[0236] The machine learning structure of the present invention embedded with the boiler mechanism is as follows Figure 4 As shown, Figure 1 As shown, the framework of the method of the present invention is summarized as follows:

[0237] 1) Numerical simulation of typical working conditions.

[0238] Collect the operation data of power plant boilers, mainly including air volume, coal volume, load, damper opening, smoke temperature and air temperature at the measuring point, wall heat load and other data under different loads. By analyzing the operation data of the boiler, screen out the operating data that can represent the on-site operation situation, and use the air volume, coal volume, damper opening, and coal quality analysis data as numerical simulation boundary conditions to perform numerical simulation of typical working conditions, and use the measuring point data to correct the numerical simulation calculation results to form a typical working condition database containing the numerical simulation calculation results of multiple typical working conditions.

[0239] 2) Machine learning and algorithm prediction.

[0240] Considering that boiler combustion is a multi-input and multi-output process, in order to accurately predict the changes in combustion parameters, a machine learning algorithm is used to build a prediction model for each combustion parameter. A combustion parameter and position information are extracted from the numerical simulation results of typical working conditions. The extracted data and the actual operating parameters of the boiler corresponding to the working condition are combined into training data for the machine learning algorithm. Through programming software, a machine learning scheme with fast learning speed, few training parameters and good generalization performance is selected to generate a machine learning algorithm learning machine. The numerical simulation data after data processing is divided into training set, test set and validation set. The training set is divided into input data and output data for training the learning machine for learning. The test set and validation set data are used to verify and calibrate the machine learning algorithm learning machine to ensure the efficiency and accuracy of the learning machine in predicting operating data.

[0241] 3) Embedding conservation equations into the machine learning process

[0242] The boiler operation process includes flow, heat transfer, and combustion processes, involving factors such as turbulence, chemical reactions, and radiation heat transfer, which can be converted into conservation control equations including conservation of mass, conservation of momentum, and conservation of energy. The conservation equations are expressed in a general format. At the same time, since the problem being solved is a steady-state problem, the time term is ignored, and only the three-dimensional problem of convection and diffusion terms is considered. The control equations required can be expressed as continuity equations, dynamic equations, turbulent kinetic energy equations, turbulent kinetic energy dissipation rate equations, various chemical component equations, etc. We constrain the training process of machine learning by converting the above equations into loss functions. For example: 1. The amount of C, H, O, and N elements in the fuel, air, and combustion products during boiler operation is calculated during the training process and used as constraints satisfied by mass conservation training.

[0243] 2. The calorific value of the fuel before entering the furnace, the enthalpy and sensible heat of the air, and the enthalpy increase and heat loss of the working fluid are used as constraints to be satisfied in the energy conservation training.

[0244] 3. The consistency of the speed in all directions before and after training in the boiler operation process is used as a constraint condition to be satisfied by the momentum conservation training. By adjusting the error penalty mechanism, the training results can satisfy the control equations involved in the boiler operation, further increasing the interpretability and accuracy of the machine learning training results.

[0245] IV) Application of results

[0246] Using a machine learning algorithm learning machine that has been embedded with quasi-steps, multiple sets of data such as load, air volume, coal volume, wind rate, etc. in the field data are taken as input parameters to calculate full-load boiler temperature, nitrogen oxides and other operating data.

Claims

1. A method for acquiring internal operation data of a boiler furnace, characterized in that: The method comprises: Step 1: Obtain the operating parameters of the loads of several working conditions; Step 2: Select two cross sections near the economizer outlet, with n nodes in each cross section; Step 3: Construct training data for machine learning, and form an input parameter matrix with input coordinates and operating parameters; Step 4: Normalize the training data; Step 5: Use machine learning to train and build a prediction model; Step 6: Check whether the machine learning results satisfy the law of conservation of carbon between the cross-sections and nodes, and correct the machine learning results by the law of conservation of carbon, including: Step 6.1: Take the economizer and its vicinity as the control volume, and take the inlet and outlet of the area as the cross-sections that satisfy the carbon conservation law, that is, the carbon mass at the inlet and outlet nodes is equal, and the mass of carbon entering the furnace is equal to the mass of carbon in each product; Step 6.2: Perform cross-sectional carbon conservation test to eliminate the face-to-face errors of the three carbon-containing substances; Step 6.3: Conduct nodal carbon conservation checks to eliminate point-to-point deviations of the three carbon-containing substances; Step 7: Use the prediction model corrected in step 6 to obtain the internal operation data of the boiler furnace.

2. The method for acquiring the internal operation data of a boiler furnace according to claim 1, characterized in that: The operating parameters in step 1 include: load, total air volume, total coal volume, primary air volume, secondary air volume, secondary air temperature, coal mill operation mode, coal mill swing angle, outlet flue gas mass flow rate, unburned carbon C of each grid node, B Mass flow rate, CO and CO2 mass fraction.

3. The method for acquiring the internal operation data of a boiler furnace according to claim 1, characterized in that: Step 3 includes: The input coordinates and operating parameters form an input parameter matrix (x, y, z, fuhe, TF, TA, T1A, T2A,, W, M, J), where x, y, z are the three-dimensional coordinate values, fuhe is the load, TF is the total coal volume, TA is the total air volume, T1A is the total primary air volume, T2A is the total secondary air volume, W is the temperature, M is the height of the coal mill, and J is the burnout air swing angle; The output coordinates and nitric oxide ppm, temperature, and carbon-containing component mass fraction form the output parameter matrix (x, y, z, m CB ,1e4*m CO ,m CO2 , T, NO), where T is temperature, NO is nitric oxide concentration, m CB is the unburned carbon, 1e4*m CO is carbon monoxide, m CO2 For carbon dioxide.

4. The method for acquiring the internal operation data of a boiler furnace according to claim 1, characterized in that: Step 5 includes: Step 5.1: Initialize the ELM model; Step 5.2: Add neurons. W and B are automatically randomly generated by default. Step 5.3: Train the ELM model using the labeled training set; Step 5.4: Input the test data and give the predicted value on the test set; Step 5.5: Use mean square error (MSE) and relative error (EPS) to evaluate the accuracy of the model.

5. The method for acquiring the internal operation data of a boiler furnace according to claim 1, characterized in that: Step 6.1 includes: Among them, Y CO , is the gas mass fraction, is the carbon mass flow rate at point (x,y,z), is the mass flow rate of unburned carbon, is the mass flow rate of carbon monoxide, is the mass flow rate of carbon dioxide; is the flue gas mass flow rate; The carbon conservation equation is: in, is the divergence of carbon mass flow, m' C " is the carbon source term, which is 0 for the economizer region. The equations applied to this region include: Cross-sectional carbon conservation equation: Nodal carbon conservation equation:

6. The method for acquiring the internal operation data of a boiler furnace according to claim 1, characterized in that: Step 6.2 includes: Step 6.2.1: Carry out cross-section carbon conservation test, the test criteria are: b T T test a=0 in, Step 6.2.2: Calculate the face-to-face deviation of the three carbon-containing substances, including the unburned carbon m CB , carbon monoxide 1e4*m CO and carbon dioxide m CO2 , specifically including: Use the ELM model to predict the output variable T, T_test = Model.predict(X_test); Find the deviation of the three carbon-containing substances from each other: e1 = numpy.dot(bT,m CB ), numpy is a database calling algorithm in python, dot is a function in the algorithm, which is used to sum the unburned carbon in each surface and then make the difference between the two surfaces; e2 = numpy.dot(bT,m CO ), which is used to sum the carbon monoxide in each face and then make the difference between the two faces; e3 = numpy.dot(bT,m CO2 ), which is used to sum the carbon dioxide in each surface and then make the difference between the two surfaces; Step 6.2.3: Assume that the ELM model is accurate in predicting the average value of each substance at 2n points, so the two surfaces each bear half of the correction; this half of the correction is evenly distributed among the n points in the surface: T1 is the output variable matrix after correction, T is the output variable matrix before correction, and the deviation correction algorithm for the three substances is as follows: Export: T1[:,0]=T[:,0]-1 / 2*e1 / n Entry: T1[:,0]=T[:,0]+1 / 2*e1 / n Export: T1[:,1]=T[:,1]-1 / 2*e2 / n Entry: T1[:,1]=T[:,1]+1 / 2*e2 / n Export: T1[:,2]=T[:,2]-1 / 2*e3 / n Entry: T1[:,2]=T[:,2]+1 / 2*e3 / n.

7. The method for acquiring the internal operation data of a boiler furnace according to claim 1, characterized in that: Step 6.3 includes: Step 6.3.1: Conduct node carbon conservation test, the test criteria are: in P=I 6×6 -aa(aa T aa) -1 , T y is the node temperature value on the y-axis section, T y+dy is the node temperature value on the section with ordinate y+dy; Step 6.3.2: Calculate the point-to-point deviations of the three carbon-containing substances, including: Use the ELM model to predict the output variable T, T_test = Model.predict(X_test); Construct matrix bb for point-by-point subtraction: Find the point-to-point deviations of the three carbon-containing species: e4=abs(np.dot(bb,T_test[:,0])).max(axis=0), find the n positions with the largest absolute deviation of unburned carbon. e5 = abs(np.dot(bb,T_test[:,1])).max(axis = 0), find the n positions with the largest absolute deviation of carbon monoxide, e6 = abs(np.dot(bb,T_test[:,2])).max(axis = 0), find the n positions with the largest absolute deviation of carbon dioxide; Find the maximum absolute value of the total carbon deviation at n positions: abs(np.dot(np.dot(bb,nT_test),[112 / 2812 / 44])).max(axis=0); Step 6.3.3: Eliminate the deviations found in the above steps through the conservation of carbon: Use the ELM model to predict the output T, T_test = Model.predict(X_test), where T_test is the output value of the prediction parameter, and X_test is the corresponding input parameter of the prediction parameter; Write T as an n*6-dimensional matrix TT: TT=[T y T y+dy Multiply the projection matrix P so that aa is in In the null space of: Write back as a 2n*3 dimensional matrix The matrix of node temperature values ​​after eliminating deviation.

8. A system for acquiring operating data inside a boiler furnace, characterized in that: The system comprises: An operating parameter acquisition module is used to obtain operating parameters of loads under several working conditions; The section selection module is used to select two layers of sections near the outlet of the economizer, with n nodes in each layer of section; The learning data construction module is used to construct the training data for machine learning, and to form an input parameter matrix with input coordinates and operating parameters; Normalization module, used to normalize the training data; Prediction model building module, used to use machine learning for learning and training to build a prediction model; The prediction model correction module is used to check whether the machine learning results satisfy the law of conservation of carbon between the cross-section and the node, and to correct the machine learning results by the law of conservation of carbon, including: Step 6.1: Take the economizer and its vicinity as the control volume, and take the inlet and outlet of the area as the cross-sections that satisfy the carbon conservation law, that is, the carbon mass at the inlet and outlet nodes is equal, and the mass of carbon entering the furnace is equal to the mass of carbon in each product; Step 6.2: Perform cross-sectional carbon conservation test to eliminate the face-to-face errors of the three carbon-containing substances; Step 6.3: Conduct nodal carbon conservation checks to eliminate point-to-point deviations of the three carbon-containing substances; The operation data acquisition module is used to obtain the internal operation data of the boiler furnace using the revised prediction model.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor runs the computer program stored in the memory, the steps of the method according to any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of computer instructions, and the plurality of computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images