CN113947013A - Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method - Google Patents

Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method Download PDF

Info

Publication number
CN113947013A
CN113947013A CN202111077632.4A CN202111077632A CN113947013A CN 113947013 A CN113947013 A CN 113947013A CN 202111077632 A CN202111077632 A CN 202111077632A CN 113947013 A CN113947013 A CN 113947013A
Authority
CN
China
Prior art keywords
data
boiler
time
neural network
model
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.)
Pending
Application number
CN202111077632.4A
Other languages
Chinese (zh)
Inventor
殷喆
杨春来
袁晓磊
李剑锋
侯倩
饶群
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.)
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Hebei Electric Power Co Ltd
State Grid Hebei Energy Technology Service Co Ltd
Original Assignee
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Hebei Electric Power Co Ltd
State Grid Hebei Energy Technology Service 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 State Grid Corp of China SGCC, Electric Power Research Institute of State Grid Hebei Electric Power Co Ltd, State Grid Hebei Energy Technology Service Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN202111077632.4A priority Critical patent/CN113947013A/en
Publication of CN113947013A publication Critical patent/CN113947013A/en
Pending legal-status Critical Current

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Computing Systems (AREA)
  • Computational Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Geometry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Operations Research (AREA)
  • Probability & Statistics with Applications (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention relates to boiler short-term NO based on hybrid deep neural network modelingxAn emissions prediction method comprising the steps of: (1) data preprocessing is carried out, and data missing and distortion conditions existing in the original data are supplemented and cleaned; (2) a feature extraction module is carried out, the CNN part of the feature extraction layer needs to carry out feature extraction on a high-dimensional time sequence of boiler operation, and the LSTM completes modeling and analysis on time dimensions of the extracted features; (3) for short-term machine based on mixed deep neural network modelGroup NOxEmissions are predicted. The invention dynamically and quantitatively predicts NO in a period of time in the future by training a CNN-LSTM deep network model and learning the time correlation among spatial featuresxThe variation of the emission can provide reference for the operation guidance of the boiler.

Description

Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method
Technical Field
The invention relates to the technical field of exhaust emission, in particular to boiler short-term NO based on hybrid deep neural network modelingxAn emission prediction method.
Background
The boiler is one of three major devices in a power plant, the combustion state of the boiler directly influences the efficiency of a generating set and the safety and stability of the boiler, and the combustion state in the boiler influences pollutant NOxThe amount of production of (c). In order to solve the problem of insufficient combustion of pulverized coal in a hearth, the oxygen content of flue gas and the overall temperature of the hearth need to be increased, but the increased oxygen content and high temperature cause NOxThe amount of production of (2) increases. In order to improve the combustion efficiency of the boiler and effectively reduce the pollutant discharge amount, related technicians need to optimize the combustion of the boiler, so that the economy and the environmental protection of the combustion operation of the boiler are both considered, and the primary task is to establish NO meeting the precisionxAn emissions prediction model. In the face of high-dimensional data with mass characteristics in the boiler combustion process, boiler NO is traditionally carried out based on LSTM modelingxThe problems of important information loss and model convergence incapability caused by overhigh data dimension and the like of the emission prediction method. Meanwhile, the traditional neural network needs to be predicted manually through features designed by priori knowledge, and the manually screened features can cause a large amount of information to be lost and seriously damage the relevance of historical data.
With NOxThe discharge-related data has various types and strong coupling, and the CNN structure has the neurons with learnable weight and deviation value, so that the low-level features of the data can be increased, and the low-level features are combined into multi-level features as the network deepens, so that a subsequent model is guided to learn and adjust the features. Therefore, the CNN structure enables the forward transfer function to be more efficient by compiling specific features into a convolution structure, and the number of parameters in the network can be reduced.
The LSTM model is a special form of RNN that provides feedback to each neuron. Its output depends not only on the current neuron inputs and weights, but also on previous neuron inputs, so theoretically the RNN structure is suitable for processing time series data, but when processing a long string of data samples, explosion and vanishing gradient problems arise, which become the entry point for using the LSTM model.
Disclosure of Invention
The invention provides a boiler short-term NO based on hybrid deep neural network modelingxAn emission prediction method for solving the problem that the conventional method cannot solve high dimensional data and short time group NOxAnd (4) an emission prediction problem. The method has the main idea that automatic feature extraction is carried out on more input data by utilizing a shallow CNN in the algorithm, and then time sequence modeling is carried out on features provided by the CNN by using LSTM to obtain a NOx emission prediction result.
The technical scheme of the invention is as follows:
boiler short-term NO based on hybrid deep neural network modelingxAn emissions prediction method comprising the steps of:
(1) data preprocessing is carried out, and data missing and distortion conditions existing in the original data are supplemented and cleaned;
(2) a feature extraction module is carried out, the CNN part of the feature extraction layer needs to carry out feature extraction on a high-dimensional time sequence of boiler operation, and the LSTM completes modeling and analysis on time dimensions of the extracted features;
(3) for short-term unit NO based on mixed deep neural network modelxEmissions are predicted.
Preferably, the step (1) comprises (a) selecting operation data with a time span of 12 months from an operation database of the power station boiler system, recording the operation data as a data set D, wherein the sampling frequency is 1 data sample per minute, and the boiler combustion system has no fault or shutdown process within the time span range of data acquisition;
(b) the acquired data is preprocessed, the data is supplemented and cleaned aiming at the data missing and distortion conditions of the original data, and the processed data is normalized in order to accelerate the convergence of a loss function. Preferably, the variables collected in step (a) comprise unit load, total coal quantity, total air quantity, primary air pressure, total primary air quantity, total secondary air quantity, opening degree of each layer of secondary air door, opening degree of each layer of burnout air door, coal feeding quantity of each layer of coal feeder, boiler outlet flue gas flow, boiler outlet flue gas temperature, boiler outlet flue gas oxygen content, boiler outlet NOxConcentration, SCR inlet NOx concentration, SCR ammonia injection amount, SCR outlet NOxConcentration, SCR ammonia escape amount, SCR inlet flue gas pressure, SCR outlet flue gas pressure, power supply coal consumption, power generation coal consumption and boiler efficiency.
Preferably, the step (b) includes:
step 2-1: based on normal distribution test, the numerical value is distributed on the assumption of normal distribution in statistics
Figure BDA0003261371760000021
The probability within is 0.9974, and if the sample deviates from the average value by more than 3 times of standard deviation, the sample is considered as an outlier;
step 2-2: after the outliers are deleted, in order to keep the trend of the data and not reduce the number of deep network training samples, linear interpolation is carried out for supplementing the outliers, as shown in the following formula,
Figure BDA0003261371760000022
step 2-3: carrying out normalization calculation on the data, wherein the calculation formula of the normalization calculation is as follows:
Figure BDA0003261371760000023
wherein the content of the first and second substances,
Figure BDA0003261371760000024
represents the normalized boiler operating data of the boiler parameters,
Figure BDA0003261371760000025
is the average of the operational data of the boiler,
Figure BDA0003261371760000026
ε is a non-zero constant for variance of boiler operating data.
Preferably, step (2) comprises: (c) a characteristic extraction module is carried out, wherein the CNN part of the characteristic extraction layer is used for carrying out characteristic extraction on the high-dimensional time sequence of boiler operation;
(d) according to the data format requirement of the LSTM model, carrying out data recombination on the obtained CNN output matrix to obtain a model parameter matrix XtAnd using the model parameter matrix XtAnd constructing an LSTM boiler emission prediction model for the variables.
Preferably, the step (c) includes converting the normalized data per minute into a kind of image, that is, regarding each kind of data as a feature, and each feature includes different measuring point data or operation parameters, and splicing the data acquired each time into two-bit panel data in a row, where the panel data includes spatial features, so as to facilitate feature extraction of the panel data by using a convolutional neural network in the following process.
Preferably, the convolutional neural network used in step (c) comprises a convolutional layer and a pooling layer, and global features representative of the panel data are extracted; CNN-based NOxEmission prediction feature extraction volumeThe lamination scans panel data lines by adopting a plurality of convolution kernels for the panel data input at each moment, extracts features through convolution operation, and performs pooling operation on the extracted feature lines; implementing pooling applies a maximization operation to the results of each filter; while Dropout techniques are used to improve the generalization of the model.
Preferably, the model parameter matrix X of step (d)tComprises the following steps:
Figure BDA0003261371760000031
wherein T is the time step number in the super-parameter of the initial prediction model of the discharge amount of the LSTM boiler,
Figure BDA0003261371760000032
representing the data of the boiler normalization parameter matrix obtained at the time of T- (T-alpha) after CNN data characteristic extraction, wherein the value range of the parameter alpha is more than or equal to 1 and less than or equal to alpha and less than or equal to T, ytIs an output parameter of the LSTM model and represents the NO of the unit at the time txThe amount of discharge of (c).
Preferably, after the features are extracted in step (3), the method is applied to NOxEmission with time-dependent large delay characteristics using LSTM to output NOxPredicted value of emission.
Preferably, the specific steps of step (3) include:
step 5-1: initializing the hyper-parameters of the LSTM model;
step 5-2: the mean square error of the sample data and the predicted value of the model is used as an error function, and the expression formula is as follows,
Figure BDA0003261371760000033
wherein X represents panel data at m times, h (X)(i)) The model at each moment represents a predicted value of the unit NOx emission, and Y represents an actual value of the unit NOx emission at each moment;
step 5-3: determining an activation function, and adopting a tahn activation function to control the output within the range of [0,1 ];
step 5-4: training the LSTM prediction model for each time tnData of (1), model using tnIncluding t before the timenTraining the network with the data at the time as training set to predict (t)n,tn+Tn) A unit NOx emission predicted value in a time period; after the prediction is complete, the time window is slid, using tn+1＝tn+TnTraining the network by data before the moment; repeating the above process, continuously sliding the time window until tNAt that time, the training of the LSTM model is completed.
The invention has the following beneficial effects:
complicated combustion process of boiler and NOxThe invention provides boiler short-term NO based on mixed deep neural network modeling, and the boiler short-term NO has spatial characteristics and time domain dependent characteristicsxThe emission prediction method dynamically and quantitatively predicts NO in a future period of time by training a CNN-LSTM deep network model and learning time correlation among spatial featuresxThe variation of the emission can provide reference for the operation guidance of the boiler.
Drawings
FIG. 1 is NOxPredicting a required data panel structure diagram;
FIG. 2 is CNN-based NOxAn emission prediction feature extraction map;
FIG. 3 is a diagram of the working steps of the present invention;
fig. 4 is a system configuration diagram of the present invention.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the application, its application, or uses. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
The relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present application unless specifically stated otherwise. Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description. Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate. In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not limiting. Thus, other examples of the exemplary embodiments may have different values. It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.
In the description of the present application, it is to be understood that the orientation or positional relationship indicated by the directional terms such as "front, rear, upper, lower, left, right", "lateral, vertical, horizontal" and "top, bottom", etc., are generally based on the orientation or positional relationship shown in the drawings, and are used for convenience of description and simplicity of description only, and in the case of not making a reverse description, these directional terms do not indicate and imply that the device or element being referred to must have a particular orientation or be constructed and operated in a particular orientation, and therefore, should not be considered as limiting the scope of the present application; the terms "inner and outer" refer to the inner and outer relative to the profile of the respective component itself.
Spatially relative terms, such as "above … …," "above … …," "above … …," "above," and the like, may be used herein for ease of description to describe one device or feature's spatial relationship to another device or feature as illustrated in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if a device in the figures is turned over, devices described as "above" or "on" other devices or configurations would then be oriented "below" or "under" the other devices or configurations. Thus, the exemplary term "above … …" can include both an orientation of "above … …" and "below … …". The device may be otherwise variously oriented (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly.
It should be noted that the terms "first", "second", and the like are used to define the components, and are only used for convenience of distinguishing the corresponding components, and the terms have no special meanings unless otherwise stated, and therefore, the scope of protection of the present application is not to be construed as being limited.
As shown in FIGS. 1-4, the boiler short-term NOx emission prediction method based on the hybrid deep neural network modeling is roughly divided into three steps: firstly, preprocessing data, and supplementing and cleaning the data aiming at the conditions of data loss and distortion of original data; secondly, a feature extraction module is carried out, the CNN part of the feature extraction layer needs to carry out feature extraction on a high-dimensional time sequence of boiler operation, and the LSTM completes modeling and analysis on time dimensions of the extracted features; and finally, predicting the short-term unit NOx emission based on the mixed deep neural network model. The method has important significance for short-term prediction of NOx emission of the power station boiler and design of an optimized control strategy.
The specific scheme is as follows:
step 1: and selecting operation data with the time span of 12 months from the operation database of the power station boiler system, recording the operation data as a data set D, wherein the sampling frequency is 1 data sample per minute, and the boiler combustion system has no fault or shutdown process within the time span range of the acquired data. The collected variables comprise unit load, total coal quantity, total air quantity, primary air pressure, total primary air quantity, total secondary air quantity, secondary air door opening degrees of all layers, burnout air door opening degrees of all layers, coal feeding quantity of coal feeders of all layers, boiler outlet flue gas flow, boiler outlet flue gas temperature, boiler outlet flue gas oxygen content, boiler outlet NOx concentration, SCR inlet NOx concentration, SCR ammonia injection quantity, SCR outlet NOx concentration, SCR ammonia escape quantity, SCR inlet flue gas pressure, SCR outlet flue gas pressure, power supply coal consumption, power generation coal consumption and boiler efficiency;
step 2: preprocessing the acquired data, wherein the acquired data needs to be supplemented and cleaned aiming at the data loss and distortion conditions of the original data, and the processed data needs to be normalized in order to accelerate the convergence of a loss function;
step 2-1: the basic idea of data supplement and cleaning is that some samples in the database are considered to be seriously deviated from the statistical characteristic rule presented by most data, the samples are considered to be outlier samples and need to be deleted and filled, and the patent adopts the principle of normal distribution test, and the numerical values are distributed on the assumption of normal distribution in statistics
Figure BDA0003261371760000062
The probability of being within is 0.9974, and a sample is considered to be an outlier if it deviates from the mean by more than 3 times the standard deviation.
Step 2-2: after the outliers are deleted, in order to keep the trend of the data and not reduce the number of deep network training samples, linear interpolation is carried out for supplementing the outliers, as shown in the following formula,
Figure BDA0003261371760000061
step 2-3: carrying out normalization calculation on the data, wherein the calculation formula of the normalization calculation is as follows:
Figure BDA0003261371760000071
wherein the content of the first and second substances,
Figure BDA0003261371760000072
represents the normalized boiler operating data of the boiler parameters,
Figure BDA0003261371760000073
is the average of the operational data of the boiler,
Figure BDA0003261371760000074
ε is a non-zero constant for variance of boiler operating data.
And step 3: and (4) performing a feature extraction module, wherein the CNN part of the feature extraction layer needs to perform feature extraction on a high-dimensional time sequence of boiler operation.
Step 3-1: converting the normalized data per minute into a kind of image, i.e. regarding each kind of data as a feature, each feature contains different measuring point data or operation parameters, and splicing the data collected each time into two-bit panel data according to columns, such as
As shown in fig. 1, the panel data includes spatial features, which facilitates subsequent feature extraction of the panel data by using a convolutional neural network;
step 3-2: the convolutional neural network mainly comprises a convolutional layer and a pooling layer, and global features representative of panel data are extracted. The basic process of CNN-based NOx emission prediction feature extraction is shown in fig. 2. The convolutional layer scans the panel data lines using a plurality of convolution kernels for the panel data input at each time, extracts features through convolution operation, and performs Pooling operation (Pooling) on the extracted feature lines. The most common way to achieve pooling is to apply a max operation to the results of each filter. The role of the pooling layer is to reduce the dimension and increase the receptive field, and in order to prevent information confusion, the maximum pooling is selected as a down-sampling method in the model. While Dropout techniques are used to improve the generalization of the model. Wherein the output of CNN is as shown below,
Hc＝[hC1,hC2,hC3,…,hCi-1,hCi]
where i represents the CNN output dimension.
And 4, step 4: according to the data format requirement of the LSTM model, carrying out data recombination on the obtained CNN output matrix to obtain a model parameter matrix XtAnd using the model parameter matrix XtConstructing an LSTM boiler emission prediction model for variables, wherein a model parameter matrix XtComprises the following steps:
Figure BDA0003261371760000075
wherein T is the time step number in the super-parameter of the initial prediction model of the discharge amount of the LSTM boiler,
Figure BDA0003261371760000076
representing the data of the boiler normalization parameter matrix obtained at the time of T- (T-alpha) after CNN data characteristic extraction, wherein the value range of the parameter alpha is more than or equal to 1 and less than or equal to alpha and less than or equal to T, ytAnd the output parameter of the LSTM model represents the NOx emission of the unit at the time t.
And 5: after the features are extracted, LSTM is used to output a predicted value of NOx emissions for the time-dependent large time-lapse characteristic of NOx emissions.
Step 5-1: initializing the hyper-parameters of the LSTM model;
step 5-2: the mean square error of the sample data and the predicted value of the model is used herein as an error function, expressed as follows,
Figure BDA0003261371760000081
wherein X represents panel data at m times, h (X)(i)) Model-based prediction of NOx emissions from a unit at each timeThe value, Y, represents the actual value of the unit NOx emission at each moment.
Step 5-3: determining an activation function, and adopting a tahn activation function to control the output within the range of [0,1 ];
step 5-4: training the LSTM prediction model for each time tnData of (1), model using tnIncluding t before the timenTraining the network with the data at the time as training set to predict (t)n,tn+Tn) A crew NOx emission prediction value over a period of time. After the prediction is complete, the time window is slid, using tn+1＝tn+TnThe data before the time trains the network. Repeating the above process, continuously sliding the time window until tNAt that time, the training of the LSTM model is completed.

Claims (10)

1. Boiler short-term NO based on hybrid deep neural network modelingxAn emissions prediction method, comprising the steps of: (1) data preprocessing is carried out, and data missing and distortion conditions existing in the original data are supplemented and cleaned;
(2) performing feature extraction, wherein the CNN part of the feature extraction layer needs to perform feature extraction on a high-dimensional time sequence of boiler operation, and the LSTM completes modeling and analysis on time dimensions of the extracted features; (3) for short-term unit NO based on mixed deep neural network modelxEmissions are predicted.
2. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 1xThe emission prediction method is characterized in that the step (1) comprises the steps of (a) selecting operation data with the time span of 12 months from an operation database of a power station boiler system, recording the operation data as a data set D, wherein the sampling frequency is 1 data sample per minute, and the boiler combustion system has no fault or shutdown process within the time span range of the acquired data;
(b) the acquired data is preprocessed, the data is supplemented and cleaned aiming at the data missing and distortion conditions of the original data, and the processed data is normalized in order to accelerate the convergence of a loss function.
3. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 2xThe emission prediction method is characterized in that the step (a) collects data, and variables of the collected data comprise unit load, total coal quantity, total air quantity, primary air pressure, total primary air quantity, total secondary air quantity, secondary air door opening degrees of all layers, burnout air door opening degrees of all layers, coal feeding quantity of all layers of coal feeders, boiler outlet flue gas flow, boiler outlet flue gas temperature, boiler outlet flue gas oxygen content, boiler outlet NOxConcentration, SCR inlet NOxConcentration, SCR ammonia injection amount, SCR outlet NOxConcentration, SCR ammonia escape amount, SCR inlet flue gas pressure, SCR outlet flue gas pressure, power supply coal consumption, power generation coal consumption and boiler efficiency.
4. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 2xAn emission prediction method, wherein the step (b) comprises:
step 2-1: based on normal distribution test, the numerical value is distributed on the assumption of normal distribution in statistics
Figure FDA0003261371750000011
The probability within is 0.9974, and if the sample deviates from the average value by more than 3 times of standard deviation, the sample is considered as an outlier;
step 2-2: after the outliers are deleted, in order to keep the trend of the data and not reduce the number of deep network training samples, linear interpolation is carried out for supplementing the outliers, as shown in the following formula,
Figure FDA0003261371750000012
step 2-3: carrying out normalization calculation on the data, wherein the calculation formula of the normalization calculation is as follows:
Figure FDA0003261371750000021
wherein the content of the first and second substances,
Figure FDA0003261371750000022
represents the normalized boiler operating data of the boiler parameters,
Figure FDA0003261371750000023
is the average of the operational data of the boiler,
Figure FDA0003261371750000024
ε is a non-zero constant for variance of boiler operating data.
5. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 1xThe emission prediction method, wherein step (2) comprises: (c) a characteristic extraction module is carried out, wherein the CNN part of the characteristic extraction layer is used for carrying out characteristic extraction on the high-dimensional time sequence of boiler operation;
(d) according to the data format requirement of the LSTM model, carrying out data recombination on the obtained CNN output matrix to obtain a model parameter matrix XtAnd using the model parameter matrix XtAnd constructing an LSTM boiler emission prediction model for the variables.
6. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 5xThe emission prediction method is characterized in that the step (c) comprises the steps of converting the normalized data per minute into a class of image, namely, regarding each class of data as a feature, enabling each feature to contain different measuring point data or operating parameters, splicing the data acquired each time into two-bit panel data in a row, enabling the panel data to contain spatial features, and facilitating feature extraction of the panel data by adopting a convolutional neural network subsequently.
7. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 6xThe emission prediction method is characterized in that the convolutional neural network used in the step (c) comprises a convolutional layer and a pooling layer, and global features representative of panel data are extracted; CNN-based NOxThe emission prediction characteristic extraction convolution layer scans panel data lines by adopting a plurality of convolution kernels for the panel data input at each moment, extracts characteristics through convolution operation and performs pooling operation on the extracted characteristic lines; implementing pooling applies a maximization operation to the results of each filter; while Dropout techniques are used to improve the generalization of the model.
8. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 7xAn emission prediction method, characterized in that, in the step (d), a model parameter matrix X is usedtComprises the following steps:
Figure FDA0003261371750000025
wherein T is the time step number in the super-parameter of the initial prediction model of the discharge amount of the LSTM boiler,
Figure FDA0003261371750000026
representing the data of the boiler normalization parameter matrix obtained at the time of T- (T-alpha) after CNN data characteristic extraction, wherein the value range of the parameter alpha is more than or equal to 1 and less than or equal to alpha and less than or equal to T, ytIs an output parameter of the LSTM model and represents the NO of the unit at the time txThe amount of discharge of (c).
9. Boiler short term NO based on hybrid deep neural network modeling as claimed in claim 1xAn emission prediction method characterized in that after the feature extraction in the step (3), NO is targetedxEmission with time-dependent large delay characteristics using LSTM to output NOxPredicted value of emission.
10. According to claim9 boiler short-term NO based on hybrid deep neural network modelingxThe emission prediction method is characterized in that the concrete steps of the step (3) comprise:
step 5-1: initializing the hyper-parameters of the LSTM model;
step 5-2: the mean square error of the sample data and the predicted value of the model is used as an error function, and the expression formula is as follows,
Figure FDA0003261371750000031
wherein X represents panel data at m times, h (X)(i)) Model for a unit NO representing each momentxPredicted amount of emission, Y represents actual amount of NOx emission of unit at each time, Y(m)Unit NO representing m timexActual value of the discharge amount;
step 5-3: determining an activation function, and adopting a tahn activation function to control the output within the range of [0,1 ];
step 5-4: training the LSTM prediction model for each time tnData of (1), model using tnIncluding t before the timenTraining the network with the data at the time as training set to predict (t)n,tn+Tn) Set NO in time periodxA discharge prediction value; after the prediction is complete, the time window is slid, using tn+1＝tn+TnTraining the network by data before the moment; repeating the above process, continuously sliding the time window until tNAt that time, the training of the LSTM model is completed.
CN202111077632.4A 2021-09-14 2021-09-14 Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method Pending CN113947013A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111077632.4A CN113947013A (en) 2021-09-14 2021-09-14 Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111077632.4A CN113947013A (en) 2021-09-14 2021-09-14 Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method

Publications (1)

Publication Number Publication Date
CN113947013A true CN113947013A (en) 2022-01-18

Family

ID=79328351

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111077632.4A Pending CN113947013A (en) 2021-09-14 2021-09-14 Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method

Country Status (1)

Country Link
CN (1) CN113947013A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114609986A (en) * 2022-03-16 2022-06-10 中国中材国际工程股份有限公司 Cement decomposing furnace denitration regulation and control optimization system and method based on predictive control

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114609986A (en) * 2022-03-16 2022-06-10 中国中材国际工程股份有限公司 Cement decomposing furnace denitration regulation and control optimization system and method based on predictive control

Similar Documents

Publication Publication Date Title
CN113962364B (en) Multi-factor power load prediction method based on deep learning
CN111814956B (en) Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction
CN110807554B (en) Generation method and system based on wind power/photovoltaic classical scene set
CN107315884A (en) A kind of building energy consumption modeling method based on linear regression
CN106991539B (en) Energy system optimal scheduling method and device
CN112258251B (en) Grey correlation-based integrated learning prediction method and system for electric vehicle battery replacement demand
CN108596242A (en) Power grid meteorology load forecasting method based on wavelet neural network and support vector machines
US20220365521A1 (en) Virtual simulation manufacturing platform based on automatic control
CN112381673B (en) Park electricity utilization information analysis method and device based on digital twin
CN112215428A (en) Photovoltaic power generation power prediction method and system based on error correction and fuzzy logic
CN113947013A (en) Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method
CN114648176A (en) Wind-solar power consumption optimization method based on data driving
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
Chen et al. Regional wind-photovoltaic combined power generation forecasting based on a novel multi-task learning framework and TPA-LSTM
CN116957356B (en) Scenic spot carbon neutralization management method and system based on big data
CN111259912B (en) Instrument image recognition method based on AE-SVM substation inspection robot
CN103927431A (en) Power station boiler operation state monitoring method based on pyramid time frame
Liu et al. Application of Improved Deep Learning Method in Intelligent Power System
CN113869359A (en) Modular neural network-based prediction method for nitrogen oxides in urban solid waste incineration process
CN113111588A (en) NO of gas turbineXEmission concentration prediction method and device
CN113283638A (en) Load extreme curve prediction method and system based on fusion model
CN114625831A (en) Classification evaluation feedback method for load identification of smart power grid
CN115717709B (en) Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model
Nayak et al. Multi-level statistical model for forecasting solar radiation
CN115409249A (en) Photovoltaic power day-ahead prediction method and system for time-division and multi-task learning

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