Cement clinker calcination process cement kiln NO x Methods and apparatus for concentration prediction

Abstract

This invention discloses a method and apparatus for predicting NOx concentration in cement kilns during the calcination process of cement clinker. The method includes the following steps: selecting input and output variables of a NOx concentration prediction model; real-time acquisition of input and output variable data of the NOx concentration prediction model, preprocessing the data, and constructing a training sample set; constructing a self-organizing cascaded echo state network prediction model based on the training sample set; and using the constructed self-organizing cascaded echo state network prediction model to predict NOx concentration. This invention employs a self-organizing cascaded echo state network to construct the prediction model, achieving self-organization of the number of model modules (i.e., depth), improving computational efficiency. Through the cascaded echo state network, feature information of the prediction variables is extracted step-by-step from the input data, enabling online prediction of NOx concentration in cement kilns.

Description

Technical Field

[0001] This invention relates to a method and apparatus for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, belonging to the field of intelligent control in cement production. Background Technology

[0002] NOx is one of the major air pollutants generated during the calcination of cement clinker, and cement production enterprises are the third largest source of NOx emissions in my country. To protect the environment, the Ministry of Environmental Protection has introduced stricter NOx emission standards for cement production enterprises. Real-time prediction of NOx concentration generated in cement kilns during the clinker calcination process is crucial for achieving emission standards.

[0003] Currently, the closest existing implementation to this invention is "A method for predicting NOx concentration in cement production line flue gas based on a neural network with a certain number of modules". This method includes the following steps: Step S1: Based on the NOx generation mechanism and the cement production line process, relevant variables for predicting NOx concentration are selected; Step S2: The selected variable data is downloaded from the cement enterprise database and preprocessed. The variable data is formed into a series by using a sliding window method to implicitly include the time delay features contained in the variable data in the input series of the model; Step S3: Through a combination of unsupervised and supervised training, the time delay features of the input data and the correspondence features of the output data are extracted and saved into the parameters of the DNN network to establish a NOx prediction model; Step S4: The NOx concentration value for a future period is predicted by combining historical data and NOx prediction values.

[0004] In the existing experimental schemes mentioned above, the depth of the DNN network needs to be set manually based on experience, making model training relatively complex and computationally inefficient. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a method and apparatus for predicting NOx concentration in cement kilns during the cement clinker calcination process, which enables online prediction of NOx concentration in cement kilns.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows: In a first aspect, the present invention provides a method for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, comprising the following steps: Select the input and output variables for the NOx concentration prediction model; Real-time acquisition of input and output variable data of NOx concentration prediction model, preprocessing and construction of training sample set; A prediction model for a self-organizing cascaded echo state network is constructed based on the training sample set; NOx concentration was predicted using a self-organizing cascaded echo state network prediction model.

[0007] As one possible implementation of this embodiment, the input variables of the NOx concentration prediction model include the NOx concentration feedback value of the smoke chamber. Kiln main unit current feedback Temperature of feed pipe in suspension preheater C5B Suspension preheater C5A outlet temperature feedback Suspension preheater C5B outlet temperature feedback Smoke chamber temperature Exhaust air temperature from the cooler The output variable of the NOx concentration prediction model is the NOx concentration feedback value of the smoke chamber at the next moment. ,in This is the sampling step size.

[0008] As one possible implementation of this embodiment, the real-time acquisition of input and output variable data of the NOx concentration prediction model, preprocessing, and construction of a training sample set includes: The sampling period is 5 minutes, with continuous sampling. m Input and output variable data for the NOx concentration prediction model; The sampled data is processed by removing outliers, filling in missing data, and normalizing. Based on the remaining after processing n The training sample set required to construct and train a self-organizing cascaded echo state network: (1) in, , T represents the matrix transpose operation.

[0009] As one possible implementation of this embodiment, the step of constructing a self-organizing cascaded echo state network prediction model based on the training sample set includes: Constructing the hidden layer state matrix of the echo state network module and the corresponding target output matrix ; Calculate the output weight matrix and output data of the echo state network module; Repeat the above steps until the L2 norm of the output data is less than the termination threshold to obtain the number of modules in the self-organizing cascaded echo state network.

[0010] As one possible implementation of this embodiment, the hidden layer state matrix of the constructed echo state network module... and the corresponding target output matrix ,include: Set the number of network modules for the initial echo state. And make time ; Set the first Each echo state network module at time Input signal: (2) in, For the first i Each echo state network module at time The actual output; Set the first Each echo state network module at time Target output: (3) Set the first Parameters of each echo-state network module: number of input neurons Number of output neurons: 1; Number of hidden layer neurons: 1 spectral radius and sparsity ; The network input weight matrix is ​​randomly generated in a uniform distribution over the interval (-1, 1). ; Sparsity is randomly generated in a uniform distribution on the interval (-1, 1) with a sparsity of . matrix ; Calculate matrix Maximum eigenvalue ; Calculate the first The weight matrix of the hidden layer of the echo state network module : (4) Calculate the first The hidden layer time of each echo state network module Neuron states: (5) in ; Repeat the above steps until... So far, construct the first Hidden layer state matrix of each echo state network module and the corresponding target output matrix : (6) (7) In the formula, This represents the number of cleaning steps.

[0011] As one possible implementation of this embodiment, the output weight matrix and output data of the echo state network module include: Calculate the first Output weight matrix of each echo state network module: (8) in, For regularization parameters, To and Identity matrices of the same shape This represents the inverse operation of a matrix; Calculate the first Output of each echo state network module: (9) In the formula, For the first The transpose of the output weight matrix of each echo state network module This is the transpose of the hidden layer state matrix of the echo state network module.

[0012] As one possible implementation of this embodiment, the step of obtaining the number of modules in the self-organizing cascaded echo state network until the L2 norm of the output data is less than the termination threshold includes: Calculate the first The L2 norm of the output of each echo state network module ; Repeat the calculation for the next echo state network module until... ,in Let the number of modules in the self-organizing cascaded echo state network be denoted as the termination threshold. .

[0013] As one possible implementation of this embodiment, the NOx concentration prediction using the constructed self-organizing cascaded echo state network prediction model includes: According to time The self-organizing cascaded echo state network prediction model calculates the predicted NOx concentration for the next time step: (10) In the formula, For the first k Each echo state network module at time The output; This represents the number of modules in a self-organizing cascaded echo state network.

[0014] Secondly, an embodiment of the present invention provides an apparatus for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, comprising: The variable selection module is used to select the input and output variables of the NOx concentration prediction model; The training sample set construction module is used to collect input and output variable data of the NOx concentration prediction model in real time, perform preprocessing, and construct the training sample set. The model building module is used to build a prediction model of a self-organizing cascaded echo state network based on the training sample set; The NOx concentration prediction module is used to predict NOx concentration using a constructed self-organizing cascaded echo state network prediction model.

[0015] As one possible implementation of this embodiment, the input variables of the NOx concentration prediction model include the NOx concentration feedback value of the smoke chamber. Kiln main unit current feedback Temperature of feed pipe in suspension preheater C5B Suspension preheater C5A outlet temperature feedback Suspension preheater C5B outlet temperature feedback Smoke chamber temperature Exhaust air temperature from the cooler The output variable of the NOx concentration prediction model is the NOx concentration feedback value of the smoke chamber at the next moment. ,in This is the sampling step size.

[0016] The technical solutions of the embodiments of the present invention can have the following beneficial effects: This invention employs a self-organizing cascaded echo state network to construct a prediction model, achieving self-organization of the number of model modules (i.e., depth), thus improving computational efficiency. By extracting feature information of the prediction variables from the input data step by step through the cascaded echo state network, online prediction of NOx concentration in cement kilns is realized. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, according to an exemplary embodiment. Figure 2 This is a structural diagram of an apparatus for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, according to an exemplary embodiment. Figure 3 This is a flowchart illustrating, according to an exemplary embodiment, a method for predicting NOx concentration in a cement kiln during the cement clinker calcination process using the apparatus described in this invention; Figure 4 This is a topology diagram of a self-organizing cascaded echo state network prediction model according to an exemplary embodiment; Figure 5This is a comparison chart of the predicted values ​​and the measured values ​​of NOx content in this invention; Figure 6 This is an error graph showing the difference between the predicted results and the measured NOx content of this invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments: To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0019] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, comprising the following steps: Select the input and output variables for the NOx concentration prediction model; Real-time acquisition of input and output variable data of NOx concentration prediction model, preprocessing and construction of training sample set; A prediction model for a self-organizing cascaded echo state network is constructed based on the training sample set; NOx concentration was predicted using a self-organizing cascaded echo state network prediction model.

[0020] As one possible implementation of this embodiment, the input variables of the NOx concentration prediction model include the NOx concentration feedback value of the smoke chamber. Kiln main unit current feedback Temperature of feed pipe in suspension preheater C5B Suspension preheater C5A outlet temperature feedback Suspension preheater C5B outlet temperature feedback Smoke chamber temperature Exhaust air temperature from the cooler The output variable of the NOx concentration prediction model is the NOx concentration feedback value of the smoke chamber at the next moment. ,in This is the sampling step size.

[0021] As one possible implementation of this embodiment, the real-time acquisition of input and output variable data of the NOx concentration prediction model, preprocessing, and construction of a training sample set includes: The sampling period is 5 minutes, with continuous sampling. m Input and output variable data for the NOx concentration prediction model; The sampled data is processed by removing outliers, filling in missing data, and normalizing. Based on the remaining after processing n The training sample set required to construct and train a self-organizing cascaded echo state network: (1) in, , T represents the matrix transpose operation.

[0022] As one possible implementation of this embodiment, the step of constructing a self-organizing cascaded echo state network prediction model based on the training sample set includes: Constructing the hidden layer state matrix of the echo state network module and the corresponding target output matrix ; Calculate the output weight matrix and output data of the echo state network module; Repeat the above steps until the L2 norm of the output data is less than the termination threshold to obtain the number of modules in the self-organizing cascaded echo state network.

[0023] As one possible implementation of this embodiment, the hidden layer state matrix of the constructed echo state network module... and the corresponding target output matrix ,include: Set the number of network modules for the initial echo state. And make time ; Set the first Each echo state network module at time Input signal: (2) in, For the first i Each echo state network module at time The actual output; Set the first Each echo state network module at time Target output: (3) Set the first Parameters of each echo-state network module: number of input neurons Number of output neurons: 1; Number of hidden layer neurons: 1 spectral radius and sparsity ; The network input weight matrix is ​​randomly generated in a uniform distribution over the interval (-1, 1). ; Sparsity is randomly generated in a uniform distribution on the interval (-1, 1) with a sparsity of . matrix ; Calculate matrix Maximum eigenvalue ; Calculate the first The weight matrix of the hidden layer of the echo state network module : (4) Calculate the first The hidden layer time of each echo state network module Neuron states: (5) in ; Repeat the above steps until... So far, construct the first Hidden layer state matrix of each echo state network module and the corresponding target output matrix : (6) (7) In the formula, This represents the number of cleaning steps.

[0024] As one possible implementation of this embodiment, the output weight matrix and output data of the echo state network module include: Calculate the first Output weight matrix of each echo state network module: (8) in, For regularization parameters, To and Identity matrices of the same shape This represents the inverse operation of a matrix; Calculate the first Output of each echo state network module: (9) In the formula, For the first The transpose of the output weight matrix of each echo state network module This is the transpose of the hidden layer state matrix of the echo state network module.

[0025] As one possible implementation of this embodiment, the step of obtaining the number of modules in the self-organizing cascaded echo state network until the L2 norm of the output data is less than the termination threshold includes: Calculate the first The L2 norm of the output of each echo state network module ; Repeat the calculation for the next echo state network module until... ,in Let the number of modules in the self-organizing cascaded echo state network be denoted as the termination threshold. .

[0026] As one possible implementation of this embodiment, the prediction of NOx concentration using the constructed self-organizing cascaded echo state network prediction model includes: According to time The self-organizing cascaded echo state network prediction model calculates the predicted NOx concentration for the next time step: ; (10) In the formula, For the first k Each echo state network module at time The output; This represents the number of modules in a self-organizing cascaded echo state network.

[0027] like Figure 2 As shown in the figure, an embodiment of the present invention provides an apparatus for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, comprising: The variable selection module is used to select the input and output variables of the NOx concentration prediction model; The training sample set construction module is used to collect input and output variable data of the NOx concentration prediction model in real time, perform preprocessing, and construct the training sample set. The model building module is used to build a prediction model of a self-organizing cascaded echo state network based on the training sample set; The NOx concentration prediction module is used to predict NOx concentration using a constructed self-organizing cascaded echo state network prediction model.

[0028] like Figure 3 As shown, the process of predicting NOx concentration in a cement kiln during the cement clinker calcination process using the device described in this invention is as follows: Step 1: Selection of input and output variables for the prediction model: Based on the generation mechanism of nitrogen oxides in the rotary kiln flue of a cement production line, the input variable for the prediction model is determined to be: NOx concentration feedback value in the flue. Kiln main unit current feedback Temperature of feed pipe in suspension preheater C5B Suspension preheater C5A outlet temperature feedback Suspension preheater C5B outlet temperature feedback Smoke chamber temperature Exhaust air temperature from the cooler The model output variable is the NOx concentration feedback value in the smoke chamber at the next time step. ,in This is the sampling step size.

[0029] Step 2: Data Acquisition and Preprocessing: Acquire model input and output variable data online, with a sampling period of 5 minutes, and perform continuous sampling. m After performing outlier removal, missing data imputation, and normalization preprocessing, the remaining data are... n The dataset is used to construct the training sample set required for training a self-organizing cascaded echo state network. (1) in , T represents the matrix transpose operation.

[0030] Step 3: Based on the training sample set Constructing a prediction model for a self-organizing cascaded echo state network S3.1: Set the initial number of echo state network modules ; S3.1.1: Let S3.1.2: Set the first Each echo state network module at time Input signal: (2) in, For the first i Each echo state network module at time The actual output; S3.1.3: Setting the first Each echo state network module at time Target output: (3) S3.1.4: Setting the first Parameters of each echo-state network module: number of input neurons Number of output neurons: 1; Number of hidden layer neurons: 1 spectral radius sparsity ; S3.1.5: In the interval The network input weight matrix is ​​generated randomly using a uniform distribution. ; S3.1.6: In the interval The sparsity is generated randomly with uniform distribution as follows: matrix ; S3.1.7: Calculate the matrix Maximum eigenvalue ; S3.1.8: Calculate the first... The weight matrix of the hidden layer of the echo state network module : (4) S3.1.9: Calculate the first... The hidden layer time of each echo state network module Neuron states: (5) in ; S3.1.10: Update ; S3.1.11: Repeat S3.1.2 to S3.1.10 until... ; S3.2 Set the number of cleaning steps , construct the first Hidden layer state matrix of each echo state network module and the corresponding target output matrix : (6) (7) S3.3: Calculate the first Output weight matrix of each echo state network module: (8) in For regularization parameters, To and Identity matrices of the same shape This represents the inverse operation of a matrix; S3.4: Calculate the first... Output of each echo state network module: (9) S3.5: Calculate the L2 norm ; S3.6: Update ; S3.7: Repeat S3.1.1 to S3.6 until... ,in Let the number of modules in the self-organizing cascaded echo state network be denoted as the termination threshold. .

[0031] Step 4: Calculation Predicted NOx concentration at time: (10) This invention example uses 6000 sets of data from a cement production line's DCS system. The first 5000 sets are used for training, and the last 1000 sets are used for testing. The first 100 sets are then discarded. The test results are as follows: Figure 5 X-axis: Samples (per 5 minutes), Y-axis: NOx concentration (mg / m³) 3 (The dashed line represents the predicted result, and the solid line represents the measured NOx concentration; the comparison between the model prediction result and the measured NOx concentration and the error) Figure 6 As shown, the X-axis represents the number of samples (per 5 minutes), and the Y-axis represents the difference between the model prediction and the measured NOx concentration. The results demonstrate the effectiveness of this invention.

[0032] This invention employs a self-organizing cascaded echo state network to construct a prediction model, achieving self-organization of the number of model modules (i.e., depth). Through the cascaded echo state network, feature information of the predictor variables is extracted step-by-step from the input data, improving not only the computational efficiency of the prediction model but also enabling online prediction of NOx concentration in cement kilns. Compared with current measurement techniques, the prediction model of this invention is simple, easy to execute, computationally intensive, and highly efficient.

[0033] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, characterized in that, Includes the following steps: Select the input and output variables for the NOx concentration prediction model; Real-time acquisition of input and output variable data of NOx concentration prediction model, preprocessing and construction of training sample set; Constructing a self-organizing cascaded echo state network prediction model based on the training sample set: Constructing the hidden layer state matrix of the echo state network module. M k and the corresponding target output matrix D k ; Calculate the output weight matrix and output data of the echo state network module; Repeat the above steps until the L2 norm of the output data is less than the termination threshold to obtain the number of modules in the self-organizing cascaded echo state network; NOx concentration was predicted using a self-organizing cascaded echo state network prediction model.

2. The method for predicting NOx concentration in a cement kiln during the cement clinker calcination process according to claim 1, characterized in that, The input variables of the NOx concentration prediction model include the NOx concentration feedback value in the smoke chamber. u 1( t ), kiln main unit current feedback u 2( t ), C5B feed pipe temperature of suspension preheater u 3( t ), C5A suspension preheater outlet temperature feedback u 4( t ), C5B suspension preheater outlet temperature feedback u 5( t Smoke chamber temperature u 6( t ), exhaust air temperature from the cooler u 7( t The output variable of the NOx concentration prediction model is the NOx concentration feedback value of the smoke chamber at the next moment. u 1( t +1), where t This is the sampling step size.

3. The method for predicting NOx concentration in a cement kiln during the cement clinker calcination process according to claim 2, characterized in that, The real-time acquisition of input and output variable data for the NOx concentration prediction model, preprocessing, and construction of a training sample set include: The sampling period is 5 minutes, with continuous sampling. m Input and output variable data for the NOx concentration prediction model; The sampled data is processed by removing outliers, filling in missing data, and normalizing. Based on the remaining after processing n The training sample set required to construct and train a self-organizing cascaded echo state network: Oh={( u ( t ), d ( t ))| t =1,2,…, n -1}(1) in, u ( t )=[ u 1( t ), u 2( t ), u 3( t ), u 4( t ), u 5( t ), u 6( t ), u 7( t )] T , d ( t )= u 1( t +1), where T is the matrix transpose operation.

4. The method for predicting NOx concentration in a cement kiln during the cement clinker calcination process according to claim 1, characterized in that, The hidden layer state matrix of the constructed echo state network module M k and the corresponding target output matrix D k ,include: Set the number of network modules for the initial echo state. k =1, and let time =1. t =1; Set the first k Each echo state network module at time t Input signal: （2） in, o i ( t ) is the first i Each echo state network module at time t The actual output, i =1,2,…, k -1; Set the first k Each echo state network module at time t Target output: （3） Set the first k Parameters of each echo-state network module: number of input neurons V k =7+ k Number of output neurons: 1; Number of hidden layer neurons: 1 N k spectral radius ρ k and sparsity SP k ; The network input weight matrix is ​​randomly generated in a uniform distribution on the interval (-1, 1). ; Sparsity is randomly generated in a uniform distribution on the interval (-1, 1) with a sparsity of . SP k matrix ; Calculate the first k The weight matrix of the hidden layer of the echo state network module W k res : （4） Calculate the first k The hidden layer time of each echo state network module t Neuron states: （5） in S k (0) = 0; Repeat the above steps until... t = n Up to -1, construct the first k Hidden layer state matrix of each echo state network module M k and the corresponding target output matrix D k : （6） （7） In the formula, wp This represents the number of cleaning steps.

5. The method for predicting NOx concentration in a cement kiln during the cement clinker calcination process according to claim 4, characterized in that, The output weight matrix and output data of the computational echo state network module include: Calculate the first Output weight matrix of each echo state network module: （8） in, For regularization parameters, To and Identity matrices of the same shape This represents the inverse operation of a matrix; Calculate the first Output of each echo state network module: （9） In the formula, For the first The transpose of the output weight matrix of each echo state network module This is the transpose of the hidden layer state matrix of the echo state network module.

6. The method for predicting NOx concentration in a cement kiln during the cement clinker calcination process according to claim 5, characterized in that, The process of obtaining the number of modules in the self-organizing cascaded echo state network until the L2 norm of the output data is less than the termination threshold includes: Calculate the first The L2 norm of the output of each echo state network module ; Repeat the calculation for the next echo state network module until... ,in Let the number of modules in the self-organizing cascaded echo state network be denoted as the termination threshold. .

7. The method for predicting NOx concentration in a cement kiln during the cement clinker calcination process according to claim 6, characterized in that, The method of predicting NOx concentration using the constructed self-organizing cascaded echo state network prediction model includes: According to time The self-organizing cascaded echo state network prediction model calculates the predicted NOx concentration for the next time step: （10） In the formula, For the first k Each echo state network module at time The output; This represents the number of modules in a self-organizing cascaded echo state network.

8. A device for predicting NOx concentration in a cement kiln during the calcination process of cement clinker, characterized in that, include: The variable selection module is used to select the input and output variables of the NOx concentration prediction model; The training sample set construction module is used to collect input and output variable data of the NOx concentration prediction model in real time, perform preprocessing, and construct the training sample set. The model building module is used to construct a self-organizing cascaded echo state network prediction model based on the training sample set: it constructs the hidden layer state matrix of the echo state network module. M k and the corresponding target output matrix D k ; Calculate the output weight matrix and output data of the echo state network module; Repeat the above steps until the L2 norm of the output data is less than the termination threshold to obtain the number of modules in the self-organizing cascaded echo state network; The NOx concentration prediction module is used to predict NOx concentration using a constructed self-organizing cascaded echo state network prediction model.

9. The device for predicting NOx concentration in cement kilns during the cement clinker calcination process according to claim 8, characterized in that, The input variables of the NOx concentration prediction model include the NOx concentration feedback value in the smoke chamber. u 1( t ), kiln main unit current feedback u 2( t ), C5B feed pipe temperature of suspension preheater u 3( t ), C5A suspension preheater outlet temperature feedback u 4( t ), C5B suspension preheater outlet temperature feedback u 5( t Smoke chamber temperature u 6( t ), exhaust air temperature from the cooler u 7( t The output variable of the NOx concentration prediction model is the NOx concentration feedback value of the smoke chamber at the next moment. u 1( t +1), where t This is the sampling step size.

Images