CN112884213B - Coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning - Google Patents
Coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning Download PDFInfo
- Publication number
- CN112884213B CN112884213B CN202110143854.5A CN202110143854A CN112884213B CN 112884213 B CN112884213 B CN 112884213B CN 202110143854 A CN202110143854 A CN 202110143854A CN 112884213 B CN112884213 B CN 112884213B
- Authority
- CN
- China
- Prior art keywords
- nox emission
- coal
- model
- nox
- component
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 20
- 238000013135 deep learning Methods 0.000 title claims abstract description 15
- 230000009466 transformation Effects 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 21
- 239000003245 coal Substances 0.000 claims description 8
- 230000004927 fusion Effects 0.000 claims 1
- 230000008569 process Effects 0.000 abstract description 9
- 239000000306 component Substances 0.000 description 52
- 238000013527 convolutional neural network Methods 0.000 description 52
- 230000006870 function Effects 0.000 description 18
- 238000013528 artificial neural network Methods 0.000 description 12
- 230000004913 activation Effects 0.000 description 10
- 238000002485 combustion reaction Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 9
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 description 8
- 230000000694 effects Effects 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 230000003068 static effect Effects 0.000 description 6
- 229910021529 ammonia Inorganic materials 0.000 description 4
- 238000002347 injection Methods 0.000 description 4
- 239000007924 injection Substances 0.000 description 4
- 230000000052 comparative effect Effects 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 230000015654 memory Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010531 catalytic reduction reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000012356 Product development Methods 0.000 description 1
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000006722 reduction reaction Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- 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"
-
- 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
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention provides a coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning, relates to the technical field of coal-fired boiler operation, and can accurately and stably predict NOx emission, and has good generalization performance and repeatability; the method comprises the following steps: s1, acquiring a value of a NOx emission sensitive parameter in a certain time period; s2, carrying out wavelet transformation on the acquired value of the NOx emission sensitive parameter to obtain a trend component and a high-frequency fluctuation component of the acquired value of the NOx emission sensitive parameter; s3, carrying out wavelet reconstruction on the trend component and the high-frequency fluctuation component; s4, adopting an LSTM model to dynamically predict the reconstructed trend component, and adopting a CNN model to dynamically predict the reconstructed high-frequency fluctuation component; s5, fusing the prediction results in the S4 to obtain a final NOx emission prediction result. The technical scheme provided by the invention is suitable for the process of predicting the NOx emission of the coal-fired boiler.
Description
Technical Field
The invention relates to the technical field of coal-fired boiler operation, in particular to a coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning.
Background
In recent years, with the rapid development of new energy power generation scale, the competition in aspects of long-term and continuous growth of the total capacity of the electric installation in China and the like, the average annual utilization hour number of the national thermal power generating unit is continuously reduced. In order to meet the requirements of reducing the utilization hours and frequent peak shaving, most thermal power generating units are in a variable load or even low load running state for a long time. During low-load operation, various energy efficiency indexes of the unit are reduced, the NOx emission concentration is increased due to the increase of the air quantity corresponding to unit fuel, and energy conservation and emission reduction are still challenging.
The power station boiler NOx emission prediction is the basis of combustion optimization and ammonia injection operation of a Selective Catalytic Reduction (SCR) system, is also an important technology for achieving economy and low emission of the boiler, and has been paid attention to by researchers in recent years.
Most of the combustion optimization modeling methods and formed product development at present are steady-state models established based on steady-state data of a boiler combustion system, and the prediction of the combustion optimization models under the condition of dynamic variable load is difficult to meet. In order to achieve the training and prediction of current power plant frequent load fluctuation pollutant emission models, many researchers have begun to try dynamic predictive models. Dynamic models always have a higher prediction accuracy than static models because dynamic prediction models take into account the time-variability of plant operation, adding useful information provided by the changing characteristics in the process history data. The dynamic prediction model can well solve the problem of the variability of the internal information of the frequent load of the power plant, and has good advantages in the aspect of model prediction.
Although the above models add dynamic information of data and achieve better prediction effect than static models, most models do not consider the existence of suspected noise of complex characteristic information, which may be caused by the influence of various factors in the process of frequent fluctuation and acquisition of original data, and even directly remove the so-called noise. For power station combustion data with frequent peak shaving, complex characteristics such as non-stationarity, nonlinearity and high volatility are usually included, and these characteristics are usually the most important components in the original data and should be considered. Moreover, considering the defects of easy over fitting, high convergence speed, tiny local part and the like of machine learning algorithms such as a neural network and the like, the accuracy of the prediction model is limited.
Accordingly, there is a need to develop a coal-fired boiler NOx prediction method for wavelet decomposition and dynamic hybrid deep learning to address the deficiencies of the prior art and solve or mitigate one or more of the problems described above.
Disclosure of Invention
In order to meet the peak shaving requirements of the power grid, most thermal power generating units often operate under variable load or even low load, and the NOx emission is increased. Selective Catalytic Reduction (SCR) denitration system ammonia injection is the main way to reduce NOx concentration in utility boilers. In order to improve the control quality and economy of the SCR denitration system, an accurate dynamic model of the SCR denitration reactor inlet NOx generation needs to be established. In view of the above, the invention provides a coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning, which has accurate and stable modeling and prediction effects, and compared with other typical prediction methods, the model has better generalization performance and higher repeatability and stability, and provides better choices for further realizing accurate ammonia injection and combustion optimization.
In one aspect, the invention provides a method for predicting NOx of a coal-fired boiler by wavelet decomposition and dynamic mixed deep learning, which is characterized by comprising the following steps:
S1, collecting a value of a NOx emission sensitive parameter in a certain time period (the NOx emission sensitive parameter is a related measurement parameter for predicting NOx emission, such as a load, an opening degree of a damper, primary air pressure, secondary air pressure, air quantity, coal quantity, an air distribution mode, a coal mill operation combination mode and the like);
S2, carrying out wavelet transformation on all acquired values of each NOx emission sensitive parameter to obtain a trend component and a high-frequency fluctuation component of the NOx emission sensitive parameter in the time period;
S3, carrying out wavelet reconstruction on the trend component and the high-frequency fluctuation component;
s4, adopting an LSTM model to dynamically predict the reconstructed trend component, and adopting a CNN model to dynamically predict the reconstructed high-frequency fluctuation component;
S5, fusing the prediction results obtained by the LSTM model and the CNN model to obtain a final NOx emission prediction result.
In accordance with aspects and any of the possible implementations described above, there is further provided an implementation in which each of the NOx emission-sensitive parameters is decomposed into a trend component and a plurality of high frequency fluctuation components.
In the aspects and any possible implementation manners as described above, there is further provided an implementation manner, where the content fused in step S5 includes: the prediction result of the trend component is compensated with the prediction result of the high frequency fluctuation component.
In the aspect and any possible implementation manner as described above, there is further provided an implementation manner, and the specific content of wavelet reconstruction of the trend component and the high-frequency fluctuation component in step S3 is sequence reconstruction.
In the aspect and any possible implementation manner as described above, there is further provided an implementation manner, where the basis for performing the sequence reconstruction on the high-frequency fluctuation component is the fluctuation frequency.
In the aspect and any possible implementation manner as described above, there is further provided an implementation manner, where the period of time for collecting the data in step S1 is one week, and the collection frequency is 1 time/minute.
In another aspect, the invention provides a method for modeling NOx in a coal-fired boiler by wavelet decomposition and dynamic hybrid deep learning, which is characterized by comprising the steps of:
S1, collecting historical data of NOx emission sensitive parameters and corresponding NOx emission amount;
s2, carrying out wavelet transformation on historical data of the NOx emission sensitive parameters and corresponding NOx emission amounts to obtain trend components and high-frequency fluctuation components of the NOx emission sensitive parameters and the NOx emission amounts;
S3, carrying out wavelet reconstruction on the trend component and the high-frequency fluctuation component;
S4, training the LSTM model by adopting the reconstructed trend component, and training the CNN model by adopting the reconstructed high-frequency fluctuation component to obtain a trained LSTM model and a trained CNN model;
repeating the steps S1-S4 to obtain the final model.
Aspects and any one of the possible implementations described above, further providing an implementation, where the NOx emission-sensitive parameters include load, damper opening, primary air pressure, secondary air pressure, air volume, coal volume, air distribution, and a combination of coal mill operation.
In aspects and any one of the possible implementations described above, there is further provided an implementation in which the setting parameters of the original LSTM model and the CNN model are obtained from empirical data.
In aspects and any one of the possible implementations as described above, there is further provided an implementation, the parameter setting of the trained LSTM model includes: the number of hidden layer nodes is 128, maxEpochs is 128, miniBatchSize is 8, initialnalarnRate is 0.006, dropout is 0.7;
The structure of the final CNN model includes: 1 input layer, 2 convolution layers (C1, C3), 2 downsampling layers (S2, S4), 2 full connection layers (C5, C6), 1 output layer; the input layer is 9*9 in size, the C1 layer obtains 4 characteristic surfaces with the size of 8 by using a convolution kernel with the sliding step length of 2x 2, after downsampling, the C3 layer obtains 8 characteristic surfaces with the size of 2x 2 by using a convolution kernel with the sliding step length of 3*3, after downsampling, the C3 layer obtains 8 characteristic surfaces with the size of 8 1*1, finally, the C5 layer is 32 layers and the C6 layer is 4 layers which are respectively connected with the previous layer through two full-connection layers, the layers integrate various local characteristics extracted in the previous stage, and finally, a predicted value is obtained through the output layer.
Compared with the prior art, the invention can obtain the following technical effects: the model based on wavelet decomposition and dynamic mixed deep learning, namely a WT-CNN-LSTM model, has the capability of adapting to working condition changes, has higher prediction precision and stability compared with other prediction methods, and can be further used as the basis of combustion optimization of a coal-fired boiler and accurate ammonia injection optimization of an SCR denitration system; meanwhile, the model has high-quality operation data and necessary data pretreatment, and has good application potential on boilers similar to coal-fired power plants.
Of course, it is not necessary for any of the products embodying the invention to achieve all of the technical effects described above at the same time.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of the basic structure of an LSTM neural network according to one embodiment of the present invention;
FIG. 2 is a flow chart of a dynamic prediction method for NOx emission prediction of a coal-fired boiler with mixed wavelet transforms, LSTM and CNN provided by one embodiment of the present invention;
Fig. 3 is a schematic structural diagram of a CNN module in a prediction model according to an embodiment of the present invention;
FIG. 4 is a graph comparing training data sets of LSTM-CNN All and LSTM-CNN Low provided by one embodiment of the invention;
FIG. 5 is a graph comparing LSTM-CNN All and LSTM-CNN Low test datasets provided in one embodiment of the invention;
FIG. 6 is a comparative graph of the RMSE of LSTM-CNN All and LSTM-CNN Low provided by one embodiment of the invention;
FIG. 7 is a graph of the RMSE standard deviation S comparison of LSTM-CNN All and LSTM-CNN Low provided by one embodiment of the invention;
FIG. 8 is a comparison of predictive performance of model training data sets provided by one embodiment of the invention;
FIG. 9 is a comparison of predictive performance of model test datasets provided by one embodiment of the invention;
FIG. 10 is a comparative graph of RMSE for each model provided by one embodiment of the invention;
FIG. 11 is a graph of a comparison of the standard deviation of the RMSE for each model provided by one embodiment of the invention.
Detailed Description
For a better understanding of the technical solution of the present invention, the following detailed description of the embodiments of the present invention refers to the accompanying drawings.
It should be understood that the described embodiments are merely some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
In order to capture the complex information contained in the utility boiler combustion data, the present invention utilizes signal processing techniques of wavelet decomposition transformation (WT) to decompose the raw combustion data sample into a smooth approximation component and a series of detail components. These decomposed sub-series components can be predicted from their characteristics by using a suitable time series deep neural network model, respectively, i.e., a hybrid deep learning algorithm to predict NOx emissions before entry into the SCR denitration system. Different modes in the decomposed data are captured by utilizing the characteristics of the mixed depth neural network model, so that the accuracy of the final prediction model can be effectively improved.
The invention provides a dynamic prediction method for predicting NOx emission of a coal-fired boiler by mixing wavelet transformation, LSTM and CNN. First, the original operation data is decomposed into a high frequency component and a low frequency component using wavelet transform. The high-frequency component reflects the fluctuation of the data, namely the high-frequency fluctuation component; the low frequency component reflects a trend, namely a trend component; and carrying out wavelet reconstruction on the high-frequency component and the low-frequency component to obtain the reconstructed high-frequency component and low-frequency component which are applied to subsequent training or prediction. Secondly, a deep learning long and short term memory neural network (LSTM) is utilized to dynamically model the low-frequency component, and a Convolutional Neural Network (CNN) is utilized to dynamically model the high-frequency component. And then fusing the two prediction models to obtain a final NOx emission model.
The model may be used to predict NOx emissions after modeling. First, the raw data is decomposed into an approximation component and a plurality of detail components based on a Wavelet Transform (WT) algorithm, and time and frequency patterns in the historical operating data can be captured simultaneously. And then, dynamically capturing trend characteristics of the approximate components by using an LSTM model, dynamically capturing influence factor characteristics in the detail components by using a CNN network, and finally obtaining corresponding prediction results. And finally, fusing the predicted results obtained by each component to obtain a final predicted result of NOx emission.
1. Wavelet transformation
Wavelet Transformation (WT) is similar to the fourier transformation principle in that raw data is decomposed by a wavelet function (derived from a mother wave by translation and stretching). The wavelet transformation has better capability of filtering signals, and compared with the original signal sequence, the decomposed sub-signal sequence can better embody the variation characteristics of variables and can more accurately predict the concentration of NOx. Wavelet transforms can be divided into 2 types: continuous Wavelet Transform (CWT) and Discrete Wavelet Transform (DWT).
The expression for the continuous wavelet transform of arbitrary function y (t) is:
in the formula, the main wavelet is a wavelet mother wave, the scale parameter is a wavelet component, and the translation parameter is a wavelet component.
The value of CWT is called wavelet coefficient
The expression of the discrete wavelet transform is:
wherein m is a scale parameter (decomposition level); n is a translation constant and is an integer.
The invention uses the quick discrete wavelet transform algorithm developed by Mallat, and the algorithm replaces the father wavelet and the mother wavelet with the low-pass filter and the high-pass filter. The low-pass filter is called a scale function for analyzing the low-frequency components, and the high-pass filter is called a wavelet function for analyzing the high-frequency components, and the original signal is decomposed into a plurality of sets of time sequences by the filter, wherein one set is a smooth time sequence of response trend characteristics, and the other sets are time sequences of response disturbance signal details. The selection of the mother wave has an important influence on the result.
2. LSTM neural network
Long-short term memory artificial neural network (LSTM) (Long-Short Term Memory) is a modified time-cycled neural network (Recurrent Neural Network, RNN). Because the LSTM neural network comprises a time memory unit, the time sequence long-short period dependent information can be learned, and therefore the LSTM neural network is suitable for processing and predicting interval and delay events in the time sequence. The LSTM internal state mainly comprises three control gate switches formed by four related activation functions with different functions (three of the three control gate switches are a sigmoid function and one of the three control gate switches is a forgetting gate, an input gate and an output gate, and the input and the output of information are controlled through the threshold structure, so that the forward and backward propagation of model training errors are realized, the purpose of model convergence is finally achieved, and the basic structure of the LSTM model is shown in figure 1.
As in fig. 1, C t–1 and C t are the old and new states of the cells, which run directly on the whole chain just like a conveyor belt. In LSTM, the key is the state of the cell. Cell state information is added or deleted by a gate structure consisting of a sigmoid neural network layer and a pair-wise multiplication operation (sigmoid is a nonlinear activation function) which can selectively determine whether information passes. For a given sample sequence x= (X 1,x2,…,xm), its forward calculation method can be expressed as:
ft＝sigmoid(wxfxt+whfht-1+bf) (3)
it＝sigmoid(wxixt+whiht-1+bi) (4)
gt＝tanh(wxgxt+whght-1+bg) (5)
ot＝sigmoid(wxo+whoht-1+bo) (6)
I, f, c, o is input gate, forget gate, cell state and output gate; w and b are respectively corresponding weight coefficient matrixes and bias items; sigma and tanh are sigmoid and hyperbolic tangent activation functions, respectively.
3. CNN network
The CNN network is a convolutional neural network, and the basic structure is composed of an input layer, a convolutional layer (convolutional layer), a pooling layer (pooling layer, also called a downsampling layer), a full-connection layer and an output layer. The essence of the CNN model is to extract the features of the input data by building multiple filters. As the network deepens, higher-level features are extracted, and finally robust features with invariance are obtained from the original data.
3.1 Convolutional layer
The convolutional layer is the core component of the convolutional neural network. A typical convolutional neural network consists of multiple convolutional layers, each of which may have multiple different convolutional kernels. Each convolution kernel can be regarded as a filter, each filter corresponds to a new feature map mapped after filtering, and each data in the same new feature map comes from the same identical filter, i.e. the weight sharing of the convolution kernel. At the convolution layer, the feature map (feature map) of the previous layer is convolved with a learnable convolution kernel, and then an activation function (Activation function) is passed to obtain the output feature map. The feature output corresponding to each input feature map in the convolutional layer can be expressed as:
In the method, in the process of the invention, Is the output of the j-th channel of convolution layer l,/>Net activation of the jth channel, called convolutional layer i, by outputting a feature map/>, for the previous layerAnd carrying out convolution summation and offset. /(I)Called the activation function, the present invention uses a sigmoid function as the activation function. M j represents calculation/>Input feature map subset,/>Is a convolution kernel matrix,/>Is the bias to the convolved feature map. For an output feature map/>Each input feature map/>Corresponding convolution kernel/>Possibly different, the symbol "×" indicates a convolution.
3.2 Downsampling layer
The downsampling layer typically follows the convolution layer, downsampling each input feature map by the following formula:
In the method, in the process of the invention, The net activation of the j-th channel, called downsampling layer l, which is output by the previous layer of feature map/>Obtained after downsampling weighting and biasing, beta is the weight coefficient of a downsampling layer,/>Is an offset term of the downsampling layer. Symbol down (>) represents a downsampling function by applying to the input feature map/>The output feature map is scaled down n times in both dimensions by dividing into a plurality of non-overlapping n x n region blocks by a sliding window method, and then summing, averaging or maximizing the values within each region block.
3.3 Full connection layer
In the fully connected network, all the two-dimensional feature maps are spliced into one-dimensional features serving as the input of the fully connected network. The output of the fully connected layer/can be obtained by weighting the inputs and by activating the response of the function:
xl＝f(ul) (13)
ul＝wlxl-1+bl (14)
Where u l is called the net activation of the full link layer l, which is weighted and biased by the previous layer output profile x l-1. w l is the weight coefficient of the fully connected network, and b l is the bias term of the fully connected layer l.
Convolutional neural network training is a back-propagation process that propagates back through an error function. The back propagation algorithm is a common supervised learning method for neural networks, with the goal of estimating network parameters from training samples and expected outputs. For the convolutional neural network, the convolutional kernel parameter k, the downsampling layer network weight beta, the full-connection layer network weight w, the bias parameters b of each layer and the like are mainly optimized.
The essence of the back propagation algorithm is to allow us to calculate the effective error for each network layer and derive therefrom a learning rule for the network parameters so that the actual network output is closer to the target. The back propagation algorithm is mainly based on a gradient descent method, network parameters are initialized to random values first, and then the network parameters are adjusted to the direction of reducing training errors through gradient descent until the network converges or the maximum iteration number is reached to stop.
4. Prediction method structure
The invention adopts db1 as the mother wave, which can well balance the wavelength and the smoothness, and the decomposition result can better reflect the change characteristics of the input variable. The wavelet is used to perform 3-layer decomposition on the input variable to more fully and meaningfully describe the NOx mass concentration model. The original data sample sequence is reconstructed by decomposition into an approximation sequence (general trend component) and three detail sequences (high frequency component). The approximation component is the low frequency component of the original sample sequence, following the trend of the signal, which is predicted using the dynamic LSTM model. The CNN model can capture the high volatility characteristics of the data, which is suitable for predicting detail sequences. This process flow is shown in fig. 2.
The prediction method comprises the following steps:
Step 1: all original operation data are decomposed into an approximate series c3 (t) and three detail series d1 (t), d2 (t) and d3 (t) through wavelet transformation;
Step 2: to replicate the original data sequence, it is important to reconstruct the approximation sequence and the detail sequence. And (3) recording corresponding values after the wavelet reconstruction of C3t, D1t, D2t and D3t as C3t, D1t, D2t and D3t, wherein the reconstruction process is inverse operation of the corresponding decomposition process. The sequence reconstructed by wavelet has less loss compared to the original data:
Xt＝C3t+D1t+D2t+D3t (15)
The meaning of decomposition is that different filtering signals are utilized to decompose into an approximate component and a series of detail components, and the effect of removing noise is achieved; and then the original signal is reversely constructed by a coefficient reconstruction method, and finally the denoising purpose is achieved on the basis of not deviating from the original signal, so that the precision is higher.
Step3: in the aspect of establishing NOx emission prediction, the LSTM model is adopted to dynamically predict C3t, and the detail series (comprising D1t, D2t and D3 t) is adopted to dynamically predict the CNN model.
Step 4: the final NOx prediction is obtained by compensating and fusing the prediction results of the series D1t, D2t, D3t with the results of A3t, and can be expressed as:
Example 1:
To verify the effectiveness of the model, the present invention collects about 7 days of operational data from the Distributed Control System (DCS) of the power plant under study. The unit load fluctuation of the selected data coverage is frequent, the sampling interval is 1 minute, the sampling point number is 11700, and the dynamic mixed LSTM and CNN neural network NOx emission model based on wavelet transformation is trained and tested. The invention utilizes the expansion structures of [ X (t), X (t-1), X (t-2), X (t-3), y (t-1) and y (t) ] to build a dynamic model of the SCR system inlet NOx. To verify the accuracy of the LSTM-CNN model, the separate low frequency built model LSTM low alone was used and the LSTM and BPNN dynamic and static KPLS models were compared without processed raw data. Model validity was verified using the above-mentioned field actual data, with the first 70% of the data set being the training dataset and the remaining 30% being the test dataset. In order to evaluate the predictive power of the method, two error measures are taken as evaluation indicators and compared with different models. The Root Mean Square Error (RMSE) is defined as follows:
where n is the number of samples, y i is the actual value, Is a predicted value.
The stability index of the model is an important basis for checking the reliability of the fitting model so as to avoid good and bad performance of the model. The standard deviation of the model error term is smaller, which can be measured by the standard deviation of the estimated error term, and the more accurate, stable and reliable the prediction model is. The standard deviation of the root mean square error is expressed as:
Model establishment and result analysis comparison:
Through debugging, the invention verifies that the important super parameters set by the LSTM model of the low-frequency part of the model are as follows: the number of hidden layer nodes is 128, maxeochs=128, minibatch size=8, initonalnearnrate=0.006, dropout=0.7. The CNN model structure of the verification model of the invention is shown in figure 3, namely 1 input layer, 2 convolution layers (C1, C3), 2 downsampling layers (S2, S4), 2 full connection layers (C5, C6) and 1 output layer; the input layer is 9*9 in size, the C1 layer obtains 4 characteristic surfaces with the size of 8 by using a convolution kernel with the sliding step length of 2x 2, after downsampling, the C3 layer obtains 8 characteristic surfaces with the size of 2x 2 by using a convolution kernel with the sliding step length of 3*3, after downsampling, the C3 layer obtains 8 characteristic surfaces with the size of 8 1*1, finally, the C5 layer is 32 layers and the C6 layer is 4 layers which are respectively connected with the previous layer through two full-connection layers, the layers integrate various local characteristics extracted in the previous stage, and finally, a predicted value is obtained through the output layer.
The training data set and the test data set of the integral part LSTM-CNN All and the low-frequency part LSTM-CNN Low of the built power station boiler SCR system inlet NOx emission prediction model are compared, as shown in fig. 4-5, and the blue line in the graph is an actual measurement value. As can be seen from the figure, both LSTM-CNN All and LSTM-CNN Low have better accuracy, but LSTM-CNN All, which is compensated by the high frequency part CNN model, is closer to the actual value, which is more accurate. To illustrate the stability of the proposed method, 15 replicates were performed under the given parameters. FIGS. 6-7 further provide specific comparative values for the RMSE and standard deviation S of LSTM-CNN All and LSTM-CNN Low trials. The average RMSE of training data of the LSTM-CNN All model is 3.48mg/Nm 3, the average RMSE of a test dataset is only 6.823.48mg/Nm 3, and the training and testing RMSE of the LSTM-CNN Low model is 5.09mg/Nm 3、8.61mg/Nm3.LSTM-CNNAll model respectively, which improves the training precision by 31.7% compared with the training precision of the LSTM-CNN Low model, and improves the testing precision by 20.8%. In addition, as can be seen from FIG. 4, the standard deviation of the LSTM-CNN All model and the standard deviation of the LSTM-CNN Low model are small, which shows that the stability of the model and the model is good. The good predictive performance of the experimental data shows that the NOx emission model based on the wavelet decomposed LSTM-CNN has a strong versatility.
FIGS. 8 and 9 are graphs comparing the WT-LSTM-CNN model with the LSTM, BPNN and static KPLS models alone. It should be noted that the same parameters are used for the LSTM alone, which is not wavelet transformed, and for the LSTM in the low frequency part of the LSTM-CNN model. The BPNN model uses the same dynamic structure, and the main parameters of the BPNN model are set as hidden layer node=128, learning rate=0.01, maximum training step number=1000, and training requirement accuracy=0.01. The static KPLS model adopts a Gaussian kernel function. For the training data set comparison graph of each model of fig. 8, each model can better fit the emission trend of NOx, but for the test data set of fig. 9, it can be clearly seen that the deviation of the KPLS model of the static model is larger, the generalization capability is not good, which may be that the KPLS algorithm is limited to small samples and steady-state data samples, and once the volatility of the data samples is larger, the accuracy is obviously reduced. 15 trials were also performed and the specific comparison values for each model are shown in figures 10-11, which are no longer given by the standard deviation of RMSE due to the larger test average RMSE of KPLS. The large standard deviation of the average RMSE of the BPNN model indicates that the network performs well and poorly when it appears, and that the stability is poor. Both LSTM-CNN and LSTM show better predictive performance, but the LSTM-CNN model is more accurate and stable.
The method for predicting the NOx of the coal-fired boiler by wavelet decomposition and dynamic mixed deep learning provided by the embodiment of the application is described in detail. The above description of embodiments is only for aiding in the understanding of the method of the present application and its core ideas; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.
Certain terms are used throughout the description and claims to refer to particular components. Those of skill in the art will appreciate that a hardware manufacturer may refer to the same component by different names. The description and claims do not take the form of an element differentiated by name, but rather by functionality. As referred to throughout the specification and claims, the terms "comprising," including, "and" includes "are intended to be interpreted as" including/comprising, but not limited to. By "substantially" is meant that within an acceptable error range, a person skilled in the art is able to solve the technical problem within a certain error range, substantially achieving the technical effect. The description hereinafter sets forth a preferred embodiment for practicing the application, but is not intended to limit the scope of the application, as the description is given for the purpose of illustrating the general principles of the application. The scope of the application is defined by the appended claims.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a product or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such product or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a commodity or system comprising such elements.
It should be understood that the term "and/or" as used herein is merely one relationship describing the association of the associated objects, meaning that there may be three relationships, e.g., a and/or B, may represent: a exists alone, A and B exist together, and B exists alone. In addition, the character "/" herein generally indicates that the front and rear associated objects are an "or" relationship.
While the foregoing description illustrates and describes the preferred embodiments of the present application, it is to be understood that the application is not limited to the forms disclosed herein, but is not to be construed as limited to other embodiments, and is capable of numerous other combinations, modifications and environments and is capable of changes or modifications within the scope of the inventive concept as expressed herein, either as a result of the foregoing teachings or as a result of the knowledge or technology of the relevant art. And that modifications and variations which do not depart from the spirit and scope of the application are intended to be within the scope of the appended claims.
Claims (4)
1. A method for predicting NOx in a coal-fired boiler by wavelet decomposition and dynamic mixed deep learning, the method comprising the steps of:
s1, acquiring values of NOx emission sensitive parameters in a certain time period, wherein each NOx emission sensitive parameter is decomposed into a trend component and a plurality of high-frequency fluctuation components; the NOx emission sensitive parameters comprise load, throttle baffle opening, primary air pressure, secondary air pressure, air quantity, coal quantity, air distribution mode and coal mill operation combination mode; wherein, the length of the time period for collecting data is one week, and the collection frequency is 1 time/minute;
s2, carrying out wavelet transformation on the acquired value of the NOx emission sensitive parameter to obtain a trend component and a high-frequency fluctuation component of the acquired value of the NOx emission sensitive parameter;
S3, carrying out wavelet reconstruction on the trend component and the high-frequency fluctuation component, wherein the specific content of the wavelet reconstruction on the trend component and the high-frequency fluctuation component is sequence reconstruction; the basis of the sequence reconstruction of the high-frequency fluctuation component is fluctuation frequency;
s4, adopting an LSTM model to dynamically predict the reconstructed trend component, and adopting a CNN model to dynamically predict the reconstructed high-frequency fluctuation component;
s5, fusing the prediction results obtained by the LSTM model and the CNN model to obtain a final NOx emission prediction result, wherein the fusion content comprises: the prediction result of the trend component is compensated with the prediction result of the high frequency fluctuation component.
2. A method for modeling NOx in a coal-fired boiler by wavelet decomposition and dynamic hybrid deep learning, the modeling method based on the method for predicting NOx in a coal-fired boiler by wavelet decomposition and dynamic hybrid deep learning according to claim 1, the steps of the modeling method comprising:
S1, collecting historical data of NOx emission sensitive parameters and corresponding NOx emission amount;
s2, carrying out wavelet transformation on historical data of the NOx emission sensitive parameters and corresponding NOx emission amounts to obtain trend components and high-frequency fluctuation components of the NOx emission sensitive parameters and the NOx emission amounts;
S3, carrying out wavelet reconstruction on the trend component and the high-frequency fluctuation component;
S4, training the LSTM model by adopting the reconstructed trend component, and training the CNN model by adopting the reconstructed high-frequency fluctuation component to obtain a trained LSTM model and a trained CNN model;
repeating the steps S1-S4 to obtain the final model.
3. The method for modeling NOx in a coal-fired boiler according to claim 2, wherein the NOx emission sensitive parameters include load, damper opening, primary air pressure, secondary air pressure, air volume, coal volume, air distribution mode and coal mill operation combination mode.
4. The method for modeling NOx in a coal-fired boiler according to claim 2, wherein the parameter settings of the trained LSTM model include: the number of hidden layer nodes is 128, maxEpochs is 128, miniBatchSize is 8, initialnalarnRate is 0.006, dropout is 0.7;
The structure of the final CNN model includes: 1 input layer, 2 convolution layers, 2 downsampling layers, 2 full connection layers and 1 output layer; the size of the input layer is 9*9, the convolution layer C1 obtains 4 characteristic surfaces with the size of 8 by using a convolution kernel with the sliding step length of 2 x 2, after downsampling, obtains characteristic surfaces with the size of 4 4*4 of the downsampling layer S2, the convolution layer C3 obtains 8 characteristic surfaces with the size of 2 x 2 by using a convolution kernel with the sliding step length of 3*3, after downsampling, obtains characteristic surfaces with the size of 8 1*1 of the downsampling layer S4, finally, full connection is carried out through a full connection layer C5 with the size of 32 layers and a full connection layer C6 with the size of 4 layers, and finally, a predicted value is obtained through the output layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110143854.5A CN112884213B (en) | 2021-02-02 | 2021-02-02 | Coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110143854.5A CN112884213B (en) | 2021-02-02 | 2021-02-02 | Coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112884213A CN112884213A (en) | 2021-06-01 |
| CN112884213B true CN112884213B (en) | 2024-04-19 |
Family
ID=76055767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110143854.5A Active CN112884213B (en) | 2021-02-02 | 2021-02-02 | Coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112884213B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114266200B (en) * | 2022-02-24 | 2022-07-05 | 山东大学 | Nitrogen dioxide concentration prediction method and system |
| CN117056731B (en) * | 2023-09-11 | 2024-09-27 | 重庆理工大学 | Ammonia nitrogen prediction method based on jet impact-negative pressure reactor flow signal |
| CN117369282B (en) * | 2023-11-17 | 2024-04-19 | 上海四方无锡锅炉工程有限公司 | Control method for adaptive hierarchical air supply and solid waste CFB boiler thereof |
| CN117977579B (en) * | 2024-03-28 | 2024-06-18 | 西华大学 | DWT-LSTM neural network-based adjustable load prediction method |
| CN118348813B (en) * | 2024-06-18 | 2024-08-20 | 日照盛泉新材料科技有限公司 | Reactor operation parameter optimization method based on machine learning |
Citations (3)
| 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 |
| CN110807577A (en) * | 2019-10-15 | 2020-02-18 | 中国石油天然气集团有限公司 | Pollution emission prediction method and device |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11442445B2 (en) * | 2017-08-02 | 2022-09-13 | Strong Force Iot Portfolio 2016, Llc | Data collection systems and methods with alternate routing of input channels |
-
2021
- 2021-02-02 CN CN202110143854.5A patent/CN112884213B/en active Active
Patent Citations (3)
| 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 |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
| CN110807577A (en) * | 2019-10-15 | 2020-02-18 | 中国石油天然气集团有限公司 | Pollution emission prediction method and device |
Non-Patent Citations (1)
| Title |
|---|
| 时间精度与空间信息对神经网络模型 预报 PM2.5 浓度的影响;邹思琳等;《北京大学学报(自然科学版)》;20200122;第56卷(第3期);第417-426页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112884213A (en) | 2021-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112884213B (en) | Coal-fired boiler NOx prediction method based on wavelet decomposition and dynamic mixed deep learning | |
| CN111860982B (en) | VMD-FCM-GRU-based wind power plant short-term wind power prediction method | |
| CN108256697B (en) | Prediction method for short-term load of power system | |
| CN111582551B (en) | Wind power plant short-term wind speed prediction method and system and electronic equipment | |
| CN114363195B (en) | Network flow prediction and early warning method for time and frequency spectrum residual convolution network | |
| Guardiola et al. | A computationally efficient Kalman filter based estimator for updating look-up tables applied to NOx estimation in diesel engines | |
| CN113687433B (en) | Bi-LSTM-based magnetotelluric signal denoising method and system | |
| CN108334987B (en) | Wavelet decomposition-neural network-based sea wave height prediction method | |
| CN116258269A (en) | Ultra-short-term load dynamic prediction method based on load characteristic decomposition | |
| CN114091766B (en) | CEEMDAN-LSTM-based space load prediction method | |
| CN108647807B (en) | River flow prediction method | |
| CN116362398A (en) | Power load prediction method based on modal decomposition and reconstruction and LSTM-MLR hybrid model | |
| CN110956312A (en) | Photovoltaic power distribution network voltage prediction method based on EMD-CNN deep neural network | |
| CN111242353A (en) | Wind power combined prediction modeling and prediction method | |
| CN113111592A (en) | Short-term wind power prediction method based on EMD-LSTM | |
| CN114546797A (en) | Cloud virtual machine aging prediction method based on time sequence decomposition | |
| CN114970946A (en) | PM2.5 pollution concentration long-term space prediction method based on deep learning model and empirical mode decomposition coupling | |
| CN114896895A (en) | Gate-controlled cyclic neural network-based denitration system inlet nitrogen oxide emission prediction method and system | |
| CN111365624A (en) | Intelligent terminal and method for detecting leakage of brine transportation pipeline | |
| Zhang et al. | Remaining useful life prediction of lithium-ion batteries based on TCN-DCN fusion model combined with IRRS filtering | |
| Wang et al. | A wind speed prediction method based on improved empirical mode decomposition and support vector machine | |
| CN111582132A (en) | Improved EEMD and PCNN-based gas leakage signal noise reduction method | |
| CN113722970B (en) | Photovoltaic power ultra-short-term online prediction method | |
| Ren et al. | Research on Noise Reduction Method of Pressure Pulsation Signal of Draft Tube of Hydropower Unit Based on ALIF‐SVD | |
| Cao et al. | A Time-series Prediction Algorithm Based on a Hybrid Model |
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 |