CN112613237A - CFB unit NOx emission concentration prediction method based on LSTM - Google Patents

CFB unit NOx emission concentration prediction method based on LSTM Download PDF

Info

Publication number
CN112613237A
CN112613237A CN202011621137.0A CN202011621137A CN112613237A CN 112613237 A CN112613237 A CN 112613237A CN 202011621137 A CN202011621137 A CN 202011621137A CN 112613237 A CN112613237 A CN 112613237A
Authority
CN
China
Prior art keywords
lstm
emission concentration
nox emission
data
cfb unit
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011621137.0A
Other languages
Chinese (zh)
Other versions
CN112613237B (en
Inventor
赵冰
王勇
刘志江
高明明
王进博
李文瑞
徐洁
巩晨
刘珊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin Development Branch Of Huadian International Power Co ltd
Original Assignee
Tianjin Development Branch Of Huadian International Power Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tianjin Development Branch Of Huadian International Power Co ltd filed Critical Tianjin Development Branch Of Huadian International Power Co ltd
Priority to CN202011621137.0A priority Critical patent/CN112613237B/en
Publication of CN112613237A publication Critical patent/CN112613237A/en
Application granted granted Critical
Publication of CN112613237B publication Critical patent/CN112613237B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Geometry (AREA)
  • General Business, Economics & Management (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention relates to a CFB unit NOx emission concentration prediction method based on LSTM, which comprises the following steps: s1, determining main influence factors of the emission concentration of nitrogen oxides of the CFB unit through a grey correlation method; s2, collecting field data and performing Gaussian smoothing on the air volume data; s3, in order to ensure the precision and speed of data training, all input and output data of LSTM are normalized, and the normalized processing interval is [ -1,1 ]; s4, establishing a CFB unit NOx emission concentration data model based on LSTM, and verifying through field data; s5 changes the output delay order in LSTM deep learning neural network, which makes the model have prediction function, thus overcoming the measurement delay caused by the reason of NOx concentration measuring point back. The method adopts a machine learning mode to model the NOx emission concentration of the fluidized bed, and has high precision and simple process; the model has a prediction property by changing the delay order of the output values in the training set, and the NOx emission concentration can be predicted in advance by 1-3 minutes.

Description

CFB unit NOx emission concentration prediction method based on LSTM
Technical Field
The invention belongs to the technical field of intelligent power generation, and particularly relates to a CFB unit NOx emission concentration prediction method based on LSTM, which predicts the NOx emission concentration of a circulating fluidized bed unit by machine learning.
Background
With the increasingly strict requirements of the national environmental protection agency on the pollutant emission of the thermal power plant, more and more pollutants emitted by the thermal power generating unit cannot reach the standard. In order to enable pollutant emission to reach the ultralow emission standard, most thermal power generating units complete ultralow emission modification, such as adding an SCR (selective catalytic reduction) denitration device and a wet desulphurization device at the tail of flue gas. In recent years, circulating fluidized bed units have been rapidly developed due to the advantages of good coal adaptability, large load regulation range, low pollutant original concentration discharge and the like. In 2013, the first 600MW supercritical circulating fluidized bed boiler in the world was put into operation in the Sichuan white horse power plant, and in 2015, the first 350MW supercritical circulating fluidized bed boiler in the world was put into operation in the Shanxi national gold power plant. By the end of 2018, the total installed capacity of the circulating fluidized bed boiler put into China reaches 82.3GW, the 660MW efficient ultra-supercritical circulating fluidized bed boiler which is currently developed is quickly put into engineering construction, and the circulating fluidized bed boiler which is expected to be built is the circulating fluidized bed boiler with the lowest emission and energy consumption level and the highest capacity and efficiency in the world.
In order to save cost and have the advantage of natural low NOx concentration emission of the circulating fluidized bed unit, an SNCR device is usually additionally arranged at the top of a hearth for denitration of the circulating fluidized bed unit. However, the mode has a great defect that the NOx measuring point cannot resist high temperature, so that the measuring point is generally installed at the position of a desulfurizing tower inlet or a chimney, the NOx concentration monitoring is relatively late, the measured value of an SNCR control system is delayed for 3-5 minutes, and the investment of the SNCR automatic control is greatly influenced, so that the ammonia injection amount or the urea amount for denitration of the circulating fluidized bed unit is manually controlled, and the experience and physical strength of operators are seriously tested. In order to improve the SNCR automatic input rate of the circulating fluidized bed unit, the emission concentration of the circulating fluidized bed unit needs to be modeled, and the delay problem caused by measuring point reasons is solved. The data modeling mode only needs to substitute input and output into the neural network for machine learning, and part of parameters of the machine learning are adjusted in the process, so that the method is simple and high in precision, but the model built by the method is not advanced and limited in engineering application. The invention adopts the LSTM neural network with the functions of memory and forgetting to carry out modeling, and carries out training again after changing the delay order output in the training after obtaining the model parameters, so that the model has the function of prediction, the prediction precision is high, and the problem of delay caused by the fact that a field measuring point is close to the back can be solved.
Disclosure of Invention
The invention provides a CFB unit NOx emission concentration prediction method based on LSTM (least squares) for solving the technical problems in the prior art, so that the delay brought by a NOx concentration measuring point is overcome.
The invention comprises the following technical scheme: a CFB unit NOx emission concentration prediction method based on LSTM comprises the following steps:
s1, determining main influence factors of the emission concentration of nitrogen oxides of the CFB unit through a grey correlation method;
s2, collecting field data, wherein the air volume data are subjected to Gaussian smoothing treatment because the air volume data are easy to mutate and cause interference to an algorithm; the main purpose of the Gaussian smoothing processing data is to eliminate the interference of abnormal data and improve the accuracy of the model.
S3, in order to ensure the precision and speed of data training, all input and output data of LSTM are normalized, the normalization processing interval is [ -1,1 ].
S4, establishing a CFB unit NOx emission concentration data model based on LSTM, and verifying through field data;
and S5, changing the output delay order in the LSTM deep learning neural network to enable the model to have a prediction function, so as to overcome the measurement delay caused by the reason that the NOx concentration measuring point is behind.
Further, the gray correlation method in step S1 mainly includes: 1) determining an analysis series, which generally comprises a reference series and a comparison series, wherein the reference series is the emission concentration of NOx, and the comparison series comprises 10 equipment parameters such as primary air volume, secondary air volume, bed temperature and the like; 2) the invention relates to a non-dimensionalized variable, which mainly aims to solve the problem that dimensions of data in a system are different and difficult to compare. 3) Calculating a correlation coefficient, wherein the larger the correlation coefficient under a certain working condition is, the larger the influence of factors in the comparison number series on the reference number series of the current working condition is; 4) calculating the average value of the correlation coefficients under all working conditions to calculate the correlation degree, and preventing abnormal data from appearing; 5) and selecting proper parameters as input parameters of the model after the relevance ranking.
Further, the input parameters of the finally established model comprise coal feeding quantity, bed temperature, primary air quantity, secondary air quantity, ammonia injection quantity and air preheater inlet oxygen quantity.
Further, the window size of the gaussian smoothing process in step S2 is 20.
Further, the model to which the normalized data is applied in step S3 is converged by a gradient descent method, so that the normalized data can improve the convergence speed of the model and improve the accuracy of the model. Chinese angelica root-barkThe formula for the normalization process is:
Figure BDA0002878394720000031
in MATLAB the code is: c ═ mapminmax (a, -1, 1);
further, the data model input based on LSTM in step S4 is the coal supply amount, bed temperature, primary air amount, secondary air amount, ammonia injection amount, and oxygen amount after gaussian smoothing and normalization processing.
Further, the training samples of the data model are 1000 groups, the input layer structure of the training set is 100 × 10(batch _ size is 100, step _ size is 10), the hidden layer is 2 LSTM layers composed of 10 neurons and 2 sense layers composed of 40 neurons, and the output layer is NOx emission concentration.
Further, the learning rate of the data model is 0.015, the number of cells is 20, Adam is adopted by the optimizer, and the learning step length is 80.
Further, in step S5, the change output delay order is 1min, 3min, and 5min, respectively, the model can be searched to predict the limit duration of the NOx emission concentration, the accuracy will start to decrease as the predicted time is longer, and the deviation exceeds the range allowed by the engineering, which indicates that the model can no longer accurately predict the NOx emission concentration.
Further, the data model in step S5 can lead 1min and 3min to make more accurate prediction of NOx concentration emission.
The invention has the advantages and positive effects that:
1. compared with the defects of complexity and limited precision of traditional mechanism modeling, the method provided by the invention adopts a machine learning mode to model the NOx emission concentration of the fluidized bed, and has the advantages of higher precision and simple process.
2. In the past, the fitting degree between the output value of the model and the actual value is increased only by improving the algorithm, the model is not predictive, so the practicability is limited, the model has the prediction property by changing the delay order of the output value in the training set, and the NOx emission concentration can be predicted in advance by 1-3 minutes.
3. The data model can predict the NOx emission concentration in advance for 1-3 minutes, and can provide scientific reference for field operation and SNCR control system design.
4. Although the machine algorithm is widely applied to modeling of the thermal power generating unit, actual application on site is quite rare, and the invention provides a new idea for application of the machine algorithm to the thermal power generating unit.
Drawings
FIG. 1 is the LSTM cell renewal process;
FIG. 2 is a modeling process for CFB unit NOx emission concentration using an LSTM neural network.
FIG. 3 is a training set versus test set effect of the LSTM neural network;
FIG. 4 is a graph of the effect of an LSTM neural network on the different predicted times of NOx emission concentration;
Detailed Description
To further clarify the disclosure of the present invention, its features and advantages, reference is made to the following examples taken in conjunction with the accompanying drawings.
Example (b): referring to fig. 1-4, a method for predicting LSTM-based NOx emission concentration of a CFB unit includes the steps of: s1, determining main influence factors of the emission concentration of nitrogen oxides of the CFB unit through a grey correlation method; s2, collecting field data, wherein the air volume data are subjected to Gaussian smoothing treatment because the air volume data are easy to mutate and cause interference to an algorithm; the main purpose of the Gaussian smooth processing data is to eliminate the interference of abnormal data and improve the accuracy of the model; the window size of the gaussian smoothing process in step S2 is 20; s3, in order to ensure the precision and speed of data training, all input and output data of LSTM are normalized, the normalized processing interval is [ -1,1 ]; s4, establishing a CFB unit NOx emission concentration data model based on LSTM, and verifying through field data; and S5, changing the output delay order in the LSTM deep learning neural network to enable the model to have a prediction function, so as to overcome the measurement delay caused by the reason that the NOx concentration measuring point is behind.
The grey correlation method in step S1 mainly includes: 1) determining an analysis series, which generally comprises a reference series and a comparison series, wherein the reference series is the emission concentration of NOx, and the comparison series comprises 10 equipment parameters such as primary air volume, secondary air volume, bed temperature and the like; 2) the invention relates to a non-dimensionalized variable, which mainly aims to solve the problem that dimensions of data in a system are different and difficult to compare. 3) Calculating a correlation coefficient, wherein the larger the correlation coefficient under a certain working condition is, the larger the influence of factors in the comparison number series on the reference number series of the current working condition is; 4) calculating the average value of the correlation coefficients under all working conditions to calculate the correlation degree, and preventing abnormal data from appearing; 5) and selecting proper parameters as input parameters of the model after the relevance ranking. And the input parameters for finally establishing the model comprise coal feeding quantity, bed temperature, primary air quantity, secondary air quantity, ammonia injection quantity and air preheater inlet oxygen quantity.
The model to which the normalized data is applied in step S3 is converged by a gradient descent method, so that the normalized data can improve the convergence speed of the model and improve the accuracy of the model. The formula of the normalization process is: in MATLAB
Figure BDA0002878394720000051
The middle code is: c ═ mapminmax (a, -1, 1).
The data model input based on LSTM in step S4 is the coal supply amount, bed temperature, primary air amount, secondary air amount, ammonia injection amount, and oxygen amount after gaussian smoothing and normalization processing. The training sample of the data model is 1000 groups, the input layer structure of the training set is 100 × 10(batch _ size is 100, step _ size is 10), the hidden layer is 2 LSTM layers composed of 10 neurons and 2 Dense layers composed of 40 neurons, and the output layer is NOx emission concentration. The data model learning rate is 0.015, the cell number is 20, Adam is adopted by an optimizer, and the learning step length is 80.
In step S5, the change output delay orders are 1min, 3min, and 5min, respectively, the model can be searched to predict the limit duration of the NOx emission concentration, the accuracy will start to decrease as the predicted time is longer, and the deviation exceeds the range allowed by the engineering, which indicates that the model can no longer accurately predict the NOx emission concentration. The data model in step S5 can lead 1min and 3min to make more accurate predictions of NOx concentration emissions.
The working principle is as follows:
1. LSTM neural network
The conventional RNN (recurrent neural network) has improved the situation that nodes between layers are not connected, and the RNN selecting the convergence mode of the gradient descent method still does not get rid of the defect of gradient explosion, so on the basis of the original RNN, the LSTM changes the updating mode of the memory cell by introducing the concepts of input gate, forgetting gate and output gate, thereby solving the problem. FIG. 1 shows the cell renewal process of LSTM, which mainly comprises three steps:
1) the forgetting gate first decides which information to discard. The method mainly outputs a numerical value between 0 and 1 through a Sigmoid function according to the output of the previous moment and the input of the current moment. A value of 1 indicates complete retention and a value of 0 indicates complete discard. The forgetting door works as follows:
ft＝simg(Wf[ht-1,xt]+bf) (1)
in the formula: wfA weight matrix of the forgetting gate; h ist-1For the output of the hidden layer at time t-1, xtInput at time t; bfIs the hidden layer offset vector.
2) The second step is to determine which information should be remembered. This step consists of two parts, one part of which calculates a candidate vector c 'by tanh'tThe other part is that the input gate determines which values to update through Sigmoid function, and the operation is as follows (2) and (3):
c′t＝tanh(Wc·[ht-1,xt]+bc) (2)
it＝simg(Wi·[ht-1,xt]+bi) (3)
wherein: c'tCandidate value for cell state update, Wc, Wi are weight matrix of cell update and memory gate, bc、biFor cell renewal andthe offset vectors of the gates are memorized.
3) Renewal of cell status, ct-1The unnecessary information is discarded, and the remaining information and ct-1 constitute a new cell state ctThe working mode is as follows:
ct＝ft*ct-1+it*c′t (4)
4) the output gate outputs, and the working mode is as follows:
ht＝ot*tanh(ct) (5)
2. circulating fluidized bed NOxLSTM modeling process of emission concentration
FIG. 2 shows the utilization of the LSTM neural network for the circulating fluidized bed unit NOxModeling process of emission concentration. Firstly, 6 selected groups of input quantity and NOx concentration emission are subjected to data processing and then substituted into a training set, and initial parameters of an LSTM model are defined. The LSTM initial parameters were: batch _ size is 10, step _ size is 100, cell _ size is 20, LSTM layer number is 1, cell number is 10, and Learning Rate (LR) is 0.015. And after the training is finished, evaluating the current model parameters by utilizing the evaluation indexes RMSE and MAPE of the training result, adopting the current parameters when the indexes are reached, and changing the model parameters to train again until the evaluation indexes meet the engineering requirements. At this time, the current model parameters are used to test the test set, and the output and the true value of the test set are compared to observe whether the test set meets the engineering standard, and the final result is shown in fig. 3.
3. Circulating fluidized bed NOxLSTM prediction process of emission concentration
Many applications of machine learning to engineering are usually stopped at step 2, and more parameters are optimized and algorithm is improved to improve the fitting degree of the output of the test set to the true value, which is very limited to help the actual engineering application, and the model is not well utilized.
Through research on the field process and the combustion theory of the circulating fluidized bed unit, the circulating fluidized bed unit is found to be a large-inertia and large-delay object. Model input for LSTMIn the method, the primary air quantity, the secondary air quantity, the coal feeding quantity, the ammonia injection quantity and the bed temperature are all reflected hearth input, the generation and the reduction of NOx are all completed in the hearth, a large number of complex chemical reaction processes are carried out in the process, the chemical reactions require about 2min, and NO is generatedxThere is still a considerable length of flue gas duct from the SNCR outlet to the point location, and NOxThe concentration measuring point also needs to sample the flue gas to a concentration analyzer, and the concentration analyzer can be used for NOxThe measurement of the concentration causes a delay, and thus the currently measured NOxThe concentration is actually determined by the input amount of 3-5 minutes before, which is used for predicting NO of the circulating fluidized bed unit by the LSTM modelxThe emission concentration provides feasibility, so that the delay caused by the reason of measuring points can be eliminated by changing the delay order output in the training set, and the invention leads the output in the training set by 1min, 3min and 5min respectively to search the predicted limit time. The results are shown in FIG. 4, and show that the LSTM model can lead NO by 1min and 3minxThe concentration discharge is accurately predicted, and the prediction precision is sharply reduced in 5min, so that the NO of the circulating fluidized bed unit can be predicted by the LSTM modelxThe predicted limit time for the emission concentration is 3 min.
While the preferred embodiments of the present invention have been illustrated and described, it will be appreciated by those skilled in the art that the foregoing embodiments are illustrative and not limiting, and that many changes may be made in the form and details of the embodiments of the invention without departing from the spirit and scope of the invention as defined in the appended claims. All falling within the scope of protection of the present invention.

Claims (10)

1. A CFB unit NOx emission concentration prediction method based on LSTM is characterized by comprising the following steps: s1, determining main influence factors of the emission concentration of nitrogen oxides of the CFB unit through a grey correlation method; s2, collecting field data and performing Gaussian smoothing on the air volume data; s3, all input and output data of LSTM are normalized, and the normalization processing interval is [ -1,1 ]; s4, establishing a CFB unit NOx emission concentration data model based on LSTM, and verifying through field data; s5 changes the output delay order in the LSTM deep learning neural network, and the data model of the NOx emission concentration has a prediction effect.
2. The method for predicting LSTM-based CFB unit NOx emission concentration according to claim 1, wherein: the grey correlation method in step S1 mainly includes the following steps: 1) determining an analysis sequence comprising a reference sequence and a comparison sequence; wherein the reference series is the emission concentration of NOx, and the comparison series is the equipment parameter; 2) carrying out equalization treatment on the dimensionless variables; 3) calculating a correlation coefficient, wherein the larger the correlation coefficient under a certain working condition is, the larger the influence of factors in the comparison number series on the reference number series of the current working condition is; 4) calculating the average value of the correlation coefficients under all working conditions to calculate the correlation degree, and preventing abnormal data from appearing; 5) and selecting proper parameters as input parameters of the model after the relevance ranking.
3. The LSTM-based CFB unit NOx emission concentration prediction method of claim 2, wherein: the input parameters comprise coal feeding quantity, bed temperature, primary air quantity, secondary air quantity, ammonia injection quantity and air preheater inlet oxygen quantity.
4. The method for predicting LSTM-based CFB unit NOx emission concentration according to claim 1, wherein: the window size of the gaussian smoothing process in step S2 is 20.
5. The method for predicting LSTM-based CFB unit NOx emission concentration according to claim 1, wherein: the formula of the normalization processing in step S3 is:
Figure FDA0002878394710000011
in MATLAB the code is: c ═ mapminmax (a, -1, 1).
6. The method for predicting LSTM-based CFB unit NOx emission concentration according to claim 1, wherein: the data model input based on LSTM in step S4 is the coal supply amount, bed temperature, primary air amount, secondary air amount, ammonia injection amount, and oxygen amount after gaussian smoothing and normalization processing.
7. The LSTM-based CFB unit NOx emission concentration prediction method of claim 6, wherein: the training sample of the data model is 1000 groups, the input layer structure of the training set is 100 × 10(batch _ size is 100, step _ size is 10), the hidden layer is 2 LSTM layers composed of 10 neurons and 2 Dense layers composed of 40 neurons, and the output layer is NOx emission concentration.
8. The LSTM-based CFB unit NOx emission concentration prediction method of claim 6, wherein: the data model learning rate is 0.015, the cell number is 20, Adam is adopted by an optimizer, and the learning step length is 80.
9. The method for predicting LSTM-based CFB unit NOx emission concentration according to claim 1, wherein: in step S5, the change output delay order is 1min, 3min, and 5min, respectively.
10. The method for predicting LSTM-based CFB unit NOx emission concentration according to claim 1, wherein: the data model in step S5 can lead 1min and 3min to make more accurate predictions of NOx concentration emissions.
CN202011621137.0A 2020-12-31 2020-12-31 CFB unit NOx emission concentration prediction method based on LSTM Active CN112613237B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011621137.0A CN112613237B (en) 2020-12-31 2020-12-31 CFB unit NOx emission concentration prediction method based on LSTM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011621137.0A CN112613237B (en) 2020-12-31 2020-12-31 CFB unit NOx emission concentration prediction method based on LSTM

Publications (2)

Publication Number Publication Date
CN112613237A true CN112613237A (en) 2021-04-06
CN112613237B CN112613237B (en) 2022-10-21

Family

ID=75249724

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011621137.0A Active CN112613237B (en) 2020-12-31 2020-12-31 CFB unit NOx emission concentration prediction method based on LSTM

Country Status (1)

Country Link
CN (1) CN112613237B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113112072A (en) * 2021-04-12 2021-07-13 上海电力大学 NOx emission content prediction method based on deep bidirectional LSTM
CN113947013A (en) * 2021-09-14 2022-01-18 国网河北省电力有限公司电力科学研究院 Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method
CN114399024A (en) * 2021-12-20 2022-04-26 淮阴工学院 Oil gas concentration big data intelligent detection system
CN115146833A (en) * 2022-06-14 2022-10-04 北京全应科技有限公司 Method for predicting generation concentration of boiler nitrogen oxide
CN115143452A (en) * 2022-05-20 2022-10-04 国家电投集团江西电力有限公司分宜发电厂 Full-load denitration control method for circulating fluidized bed boiler unit

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109165789A (en) * 2018-09-04 2019-01-08 广东电网有限责任公司 The modeling method and device of emission of NOx of boiler amount prediction model based on LSTM

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109165789A (en) * 2018-09-04 2019-01-08 广东电网有限责任公司 The modeling method and device of emission of NOx of boiler amount prediction model based on LSTM

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
戴金池等: "基于LSTM神经网络的柴油机NO_x排放预测", 《内燃机学报》 *
钱虹等: "基于深度循环神经网络的SCR烟气脱硝系统出口NO_x排放预测研究", 《热能动力工程》 *
高常乐等: "基于LSTM的烟气NO_x浓度动态软测量模型", 《热能动力工程》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113112072A (en) * 2021-04-12 2021-07-13 上海电力大学 NOx emission content prediction method based on deep bidirectional LSTM
CN113947013A (en) * 2021-09-14 2022-01-18 国网河北省电力有限公司电力科学研究院 Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method
CN114399024A (en) * 2021-12-20 2022-04-26 淮阴工学院 Oil gas concentration big data intelligent detection system
CN114399024B (en) * 2021-12-20 2023-02-03 淮阴工学院 Oil gas concentration big data intelligent detection system
CN115143452A (en) * 2022-05-20 2022-10-04 国家电投集团江西电力有限公司分宜发电厂 Full-load denitration control method for circulating fluidized bed boiler unit
CN115146833A (en) * 2022-06-14 2022-10-04 北京全应科技有限公司 Method for predicting generation concentration of boiler nitrogen oxide
CN115146833B (en) * 2022-06-14 2024-07-19 北京全应科技有限公司 Prediction method for generation concentration of nitrogen oxides of boiler

Also Published As

Publication number Publication date
CN112613237B (en) 2022-10-21

Similar Documents

Publication Publication Date Title
CN112613237B (en) CFB unit NOx emission concentration prediction method based on LSTM
CN112085277B (en) SCR denitration system prediction model optimization method based on machine learning
CN112488145B (en) NO based on intelligent methodxOnline prediction method and system
CN107016455B (en) Prediction system and method for oxygen content of flue gas at hearth outlet of circulating fluidized bed household garbage incineration boiler
Kang et al. Dynamic modeling of SCR denitration systems in coal-fired power plants based on a bi-directional long short-term memory method
CN111804146A (en) Intelligent ammonia injection control method and intelligent ammonia injection control device
CN106202946A (en) Clinker free calcium levels Forecasting Methodology based on degree of depth belief network model
CN104715142B (en) A kind of station boiler NOxDischarge dynamic soft-measuring method
CN113433911B (en) Accurate control system and method for ammonia spraying of denitration device based on accurate concentration prediction
CN102693451A (en) Method for predicting ammonia process flue gas desulphurization efficiency based on multiple parameters
CN109670625A (en) NOx emission concentration prediction method based on Unscented kalman filtering least square method supporting vector machine
CN110597070A (en) Method for identifying model parameters of thermal power generating unit system
CN109992844A (en) A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model
CN114186708A (en) Circulating fluidized bed unit SO based on PSO-ELM2Concentration prediction method
CN113112072A (en) NOx emission content prediction method based on deep bidirectional LSTM
CN113488111B (en) Ammonia injection amount optimization modeling method for SCR denitration system
CN115656439A (en) Online monitoring method for nitrogen oxide emission concentration of coal-fired unit based on transfer learning
CN118280472A (en) Thermal power unit SCR reactor nitrogen oxide concentration prediction method based on snow melt optimizer
CN117190173B (en) Optimal control method and control system for flue gas recirculation and boiler coupling system
CN113780639A (en) Urban solid waste incineration nitrogen oxide NOx emission prediction method based on multitask learning framework
He et al. Prediction of MSWI furnace temperature based on TS fuzzy neural network
CN113192569A (en) Harmful gas monitoring method based on improved particle swarm and error feedback neural network
Hu et al. Dynamic modeling framework based on automatic identification of operating conditions for sintering carbon consumption prediction
CN114740713B (en) Multi-objective optimization control method for wet flue gas desulfurization process
Liu et al. A prediction method of NOx in thermal power plants using GC-LSTM neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant