CN112308311A - Online prediction system and method for oxygen content of flue gas of thermal power plant - Google Patents
Online prediction system and method for oxygen content of flue gas of thermal power plant Download PDFInfo
- Publication number
- CN112308311A CN112308311A CN202011179665.5A CN202011179665A CN112308311A CN 112308311 A CN112308311 A CN 112308311A CN 202011179665 A CN202011179665 A CN 202011179665A CN 112308311 A CN112308311 A CN 112308311A
- Authority
- CN
- China
- Prior art keywords
- prediction
- model
- elman
- oxygen content
- flue gas
- 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
Links
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 title claims abstract description 53
- 239000001301 oxygen Substances 0.000 title claims abstract description 53
- 229910052760 oxygen Inorganic materials 0.000 title claims abstract description 53
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 title claims abstract description 46
- 239000003546 flue gas Substances 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000012937 correction Methods 0.000 claims abstract description 32
- 238000012360 testing method Methods 0.000 claims abstract description 30
- 238000005070 sampling Methods 0.000 claims description 14
- 238000004891 communication Methods 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 10
- 239000000779 smoke Substances 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 7
- 239000003245 coal Substances 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000000691 measurement method Methods 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 19
- 239000002245 particle Substances 0.000 description 16
- 239000011159 matrix material Substances 0.000 description 12
- 238000005259 measurement Methods 0.000 description 9
- 238000002485 combustion reaction Methods 0.000 description 8
- 238000012546 transfer Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- MCMNRKCIXSYSNV-UHFFFAOYSA-N Zirconium dioxide Chemical compound O=[Zr]=O MCMNRKCIXSYSNV-UHFFFAOYSA-N 0.000 description 4
- 238000006073 displacement reaction Methods 0.000 description 4
- 230000005611 electricity Effects 0.000 description 4
- 238000012886 linear function Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000008602 contraction Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004939 coking Methods 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 239000010791 domestic waste Substances 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000004056 waste incineration Methods 0.000 description 1
Images
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
- G06N20/00—Machine learning
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Software Systems (AREA)
- Marketing (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Water Supply & Treatment (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Feedback Control In General (AREA)
Abstract
The invention relates to an on-line prediction system and a prediction method for the oxygen content of flue gas of a thermal power plant. The method has the advantages that the Elman network is introduced into the PSO algorithm, iterative computation is prevented from falling into local minimum, iterative convergence speed of a training model is increased, the error feedback correction link enables the prediction model to have real-time self-correction capability in each step in online testing, frequent training for correcting the model due to errors is avoided, online prediction accuracy is guaranteed, and online prediction efficiency and robustness are improved. The problems of low precision, model mismatch in online test of an offline prediction model and the like in the conventional soft measurement method are solved, so that effective online prediction in a full working condition range is ensured.
Description
Technical Field
The invention belongs to the technical field of boiler thermal technology and computer monitoring intersection, and particularly relates to an on-line prediction system and a prediction method for the oxygen content of flue gas of a thermal power plant.
Background
In recent years, with the increase of energy conservation and emission reduction, the nation increasingly pays more attention to the related technical research of the economical and efficient operation of the thermal power plant. The boiler is used as important operation equipment of a thermal power plant, the combustion state of the boiler is evaluated in real time, the boiler is an important link for realizing online monitoring of combustion and optimization of combustion, and the boiler has great significance for safe and economic stable operation of a thermal power unit. In engineering practice, the excess air factor is used to determine the real-time combustion status of the boiler. However, since the parameter cannot be directly measured, the excess air coefficient at the corresponding moment is usually indirectly calculated by detecting the oxygen content at the outlet of the air preheater at the tail part of the boiler. And according to the above, the air quantity entering the hearth is adjusted, the combustion air-coal ratio of the boiler is optimized, and the boiler is in a high-efficiency and safe operation state. Excessive flue gas oxygen levels can result in heat loss from the flue gas and a decrease in boiler efficiency. Otherwise, coking in the boiler is caused, and the safety of the boiler operation is seriously threatened. Therefore, the real-time accurate measurement of the oxygen content of the flue gas of the thermal power plant is always a key technology which is highly valued in the power industry at home and abroad.
At present, the main methods for measuring the oxygen content of the smoke at home and abroad are a direct measurement method and a soft measurement method: (1) the direct measurement method directly measures the oxygen content at the tail part of the boiler by using a zirconia sensor, is simple to operate, but actually measures the NO existing in the environmentxAnd SOxWhen corrosive gas is easy to damage the sensor, the measurement precision of the sensor is reduced and the sensor is frequently replaced, so that the accurate measurement of the oxygen content of the flue gas is not facilitated, and the measurement cost is increased; (2) in recent years, a large number of scholars use soft measurement technology to predict the oxygen content of smoke which is difficult to directly and accurately measure by means of related auxiliary variables which are easy to measure in a measurement environment. The method comprises the steps of establishing a prediction model by an intelligent algorithm model, establishing a prediction model by mechanism analysisAnd (3) modeling by a hybrid method, namely a method combining mechanism analysis and an intelligent algorithm model. According to the CNKI, RNWS.0.2010-03-020 and LS-SVM and simplex-based soft measurement of the oxygen content of the flue gas, a thermal power plant flue gas oxygen content prediction method based on a minimum two-component support vector machine is provided, and the method has good off-line prediction precision, but the modeling process is trained, the calculated amount is large, and the model is not favorable for quick and accurate establishment and updating during on-line prediction; in Chinese patent CN2017101083999, the prediction system and method for oxygen content of flue GAs at the hearth outlet of a circulating fluidized bed domestic waste incineration boiler utilize PSO-GA-SVM to establish an off-line prediction model, so that network training iterative convergence speed is increased, and good off-line prediction accuracy is achieved. And a model updating module is added, namely the model can be retrained when the prediction error is larger during online test, so that the model has certain robustness. In actual operation, however, when the offline prediction model is directly used for online prediction, the accuracy is easily reduced due to complex working conditions and disturbance, and therefore frequent training and updating of the prediction model are caused. This can lead to a reduction in prediction efficiency, which is detrimental to accurate measurements on-line.
Disclosure of Invention
The invention discloses an online prediction system and a prediction method for the oxygen content of flue gas of a thermal power plant, which solve the problems of low precision in the existing soft measurement method, model mismatch in online test of an offline prediction model and the like.
In order to achieve the purpose, the technical scheme of the invention is as follows:
an on-line prediction system for the oxygen content in flue gas of a thermal power plant comprises an upper computer and a data communication interface, and is connected with a distributed control system of the thermal power plant. And (3) constructing and updating a flue gas oxygen content prediction model in the upper computer, and transmitting a prediction result meeting the precision requirement to a distributed control system through a data communication interface. Wherein the host computer includes:
(1) the signal acquisition module: the module is used for collecting boiler operation state parameters from a boiler distributed control system, and selecting main steam pressure, main steam flow, total air quantity, air preheater outlet flue gas temperature, total coal quantity, boiler load and blower current phase as input variables of a flue gas oxygen content prediction model by calculating Pearson correlation coefficients.
(2) A data preprocessing module: the module removes singular points and noise from the input data of the prediction model by using an improved wavelet double-layer threshold denoising method. And establishing training set data and on-line test set data through hierarchical sampling, and then carrying out normalization processing on the data, and mapping the data into the range of [ -1,1 ].
(3) An expert knowledge base: the module continuously updates the training samples by adopting a method of rolling a time window, so that the training samples are always kept in the latest state, wherein the method of rolling the time window refers to the step of backtracking the time size of L length from the current time, and the unit of L is second;
(4) an intelligent modeling module: establishing a relation model between related auxiliary variables and the oxygen content of the smoke by using an Elman neural network, introducing a PSO algorithm to optimize the weight and the threshold in the Elman model to obtain an optimal weight and an optimal threshold, endowing the optimal weight and the optimal threshold to the Elman network model, and completing construction of a PSO-Elman prediction model, wherein the specific process comprises the following steps:
4.1) Elman structure was established and trained: the input is 7 parameters, the output is 1 parameter, the Elman network structure is established to be 7-15-1, the hidden layer transfer function uses a tansig function, the output layer transfer function uses a linear function, the training function is tranlm, namely, the formula for correcting the network weight by adopting a Levenberg-Marquardt (L-M) algorithm is as follows:
ω(n+1)＝ω(n)-[JTJ+μI]-1JTe
wherein e is an error performance function, and J is a Jacobian matrix of the first derivative of the error performance function to the network weight. By obtaining the error of the training, the average absolute error is calculated
Where N is the number of training sessions.
4.2) introducing a PSO optimization algorithm to optimize all weights and thresholds in the Elman network structure, firstly defining a prediction error as a fitness value in the PSO algorithm, and calculating the formula as follows:
and performing population random initialization, wherein the population size of the particle group is set to be 40, the particle length is set to be 105, the maximum iteration number is set to be 100, the learning factor C1 is 2.3, the C2 is 1.3, the maximum speed vmax of the particle motion is 1, the maximum inertia weight wmax is 0.9, and the minimum wmin is 0.3. Then calculating the fitness of the individual, and selecting the local optimal position vector of the individual and the global optimal position vector of the population; then, the iteration times, each individual velocity vector and each individual position vector are set; then for each individual speed update: vid＝ωVid+C1r1(Pid-Xid)+C2r2(Pgd-Xid) And displacement updating: xid＝Xid+VidWherein V isidExample movement speeds; xidIs the current position of the particle; omega is an inertia factor; c1 and C2 are learning factors, and C1 is generally equal to C2 is equal to 2; r1, r2 is [0, 1]]A random number of (c); pid is the individual extremum for the ith example; pgd is a global extremum. And finally, judging whether the maximum iteration times or the global optimal position is reached, outputting the optimal solution after the maximum iteration times or the global optimal position is reached, and returning to continue the iteration. And obtaining the optimal weight and the threshold, and endowing the optimal weight and the threshold to the Elman model to complete the construction of the PSO-Elman model.
(5) An error correction feedback module: and operating a PSO-Elman flue gas oxygen content prediction model by using the online test set, calculating a prediction error of each step, and performing weighted feedback on the prediction error of each step to the next prediction output to finish feedback correction on the online prediction output. The specific method comprises the following steps:
using the prediction error e (k-1) of the k-1 second, i.e.
Weighted correction kth prediction: setting a weighting coefficient matrix h with dimension of 1 × 800, wherein a formula corresponding to the k-th second prediction output correction is as follows:
setting the initial values of all elements in the weighting coefficient matrix h to be 0.5, and the value interval to be [0.1], and then updating the algorithm principle of the weighting coefficient:
(step 1) calculation of prediction error
(step 2) calculating a weight coefficient compensation increment
(step 3) updating the weighting factor h (1, k) ═ 0.5+ delta (k)
(step 4) clipping the updated weighting coefficient, if h (1, k) is larger than or equal to 1, outputting h (1, k) 1, and if h (1, k) is smaller than or equal to 0, outputting h (1, k) 0
(step 5) k is k +1, and whether k is less than or equal to the maximum test step number is judged, if yes, the step 1 is returned to
Wherein
The mean absolute error of the model training calculated in step 2. Therefore, the prediction output of the next prediction model is corrected by using the prediction error of the previous step.
(6) A model evaluation module: and obtaining a prediction error of each step of the model, judging whether the error is more than or equal to 5%, if so, considering that the current prediction model is mismatched, returning to the intelligent modeling module, and training the model again by using the latest training set data from the expert knowledge base.
(7) A communication module: the module transmits the prediction result of the oxygen content of the flue gas meeting the precision requirement to a distributed control system of the thermal power plant.
A prediction method of an online prediction system for the oxygen content of flue gas of a thermal power plant is realized by combining neural network modeling and a feedback correction link, and comprises the following specific steps:
step 1: reading historical data from a DCS (distributed control system) of a thermal power plant, selecting relevant auxiliary variables for modeling, preprocessing the data of the auxiliary variables, and generating a training set and a test data set;
step 2: and establishing a relation model between the related auxiliary variables and the oxygen content of the smoke by using an Elman neural network, and training to further calculate the predicted average absolute error.
And step 3: and optimizing all weights and thresholds in the Elman network structure by introducing a PSO optimization algorithm to obtain optimal weights and thresholds, and endowing the optimal weights and thresholds to the Elman model to complete the construction of the PSO-Elman model.
And 4, step 4: and operating a PSO-Elman flue gas oxygen content prediction model by using the online test set, calculating a prediction error of each step, and performing weighted feedback on the prediction error of each step to the next prediction output to finish feedback correction on the online prediction output.
Further, in the step 1, correlation between other parameters in the operation historical data set and the oxygen content of the flue gas is calculated by utilizing a Pearson correlation coefficient, and the parameters higher than 0.8 are considered to have strong correlation with the oxygen content of the flue gas.
Further, in step 1, main steam pressure, main steam flow, total air volume, air preheater outlet flue gas temperature, total coal volume, boiler load and blower current are established as relevant auxiliary variables, and are used as input of a prediction model, and flue gas oxygen content is used as output of the prediction model, so that a data set for training the prediction model is formed.
Further, in step 1, a wavelet threshold denoising method is improved by setting a double-layer threshold, so that singular points and noise in input data of the prediction model are removed. Based on the traditional noise estimation and the general threshold calculation, the first threshold T is larger than the first threshold1Wavelet coefficient contraction bit of1In, i.e.
|dj(n)|＞T1Then re-estimating the noise formation T2By means of T2And carrying out soft threshold processing on the global signal, wherein a value taking function is as follows:
wherein
The wavelet coefficients after two threshold adjustments are respectively.
Further, in step 1, in order to ensure that the data characteristics of the operation in each time period of the whole day are covered, training set data are selected by a hierarchical sampling method. Because the mapping relation between the input data and the oxygen content of the smoke in the daily electricity utilization peak period is more complex, the daily data are divided into two layers (namely 7: 00-23: 00; 00: 00-07: 00 and 23: 00-24: 00) according to the daily electricity utilization peak period of 7: 00-23: 00. Since the system sampling frequency is 15S/time, and the sample size of the first layer data (7: 00-23: 00) is 3840, calculating the sampling ratio of the layer:
the sample size of the second layer data (00: 00-07: 00 and 23: 00-24: 00) is 1920, then the layer sampling ratio is calculated as:
thus, it is determined that the sampling ratios of the two layers of data are set to 2/3 and 1/3, respectively, 900 sets of data are extracted to form a training set for training the model. And a continuous 800 sets of data were selected for the test set for online verification.
Further, the Elman structure was established in step 2: the input is 7 parameters, the output is 1 parameter, the Elman network structure is established to be 7-15-1, the hidden layer transfer function uses a tansig function, the output layer transfer function uses a linear function, the training function is tranlm, namely, the formula for correcting the network weight by adopting a Levenberg-Marquardt (L-M) algorithm is as follows:
ω(n+1)＝ω(n)-[JTJ+μI]-1JTe
wherein e is an error performance function, and J is a Jacobian matrix of the first derivative of the error performance function to the network weight.
Further, in step 2, the prediction error is defined as a fitness value in the PSO algorithm, and the calculation formula is as follows:
further, in step 3, the optimal weight and threshold process of the Elman neural network for particle group optimization: firstly, randomly initializing a population, calculating the fitness of an individual, and selecting a local optimal position vector of the individual and a global optimal position vector of the population; then, the iteration times, each individual velocity vector and each individual position vector are set; then for each individual speed update: vid＝ωVid+C1r1(Pid-Xid)+C2r2(Pgd-Xid) And displacement updating: xid＝Xid+VidWherein V isidExample movement speeds; xidIs the current position of the particle; omega is an inertia factor; c1,C2For learning factors, take C1＝C2＝2；r1，r2Is [0, 1]]A random number of (c); pidThe individual extremum for the ith example; pgdIs a global extremum. And finally, judging whether the maximum iteration times or the global optimal position is reached, outputting the optimal solution after the maximum iteration times or the global optimal position is reached, and returning to continue the iteration.
Further, the feedback correction procedure in step 4 utilizes the prediction error e (k-1) of the k-1 second, i.e. the step
Weighted correction kth prediction: setting a weighting coefficient matrix h with dimension of 1 × 800, wherein a formula corresponding to the k-th second prediction output correction is as follows:
setting the initial values of all elements in the weighting coefficient matrix h to be 0.5, and the value interval to be [0.1], and then updating the algorithm principle of the weighting coefficient:
(step 1) calculation of prediction error
(step 2) calculating a weight coefficient compensation increment
(step 3) updating the weighting factor h (1, k) ═ 0.5+ delta (k)
(step 4) clipping the updated weighting coefficient, if h (1, k) is larger than or equal to 1, outputting h (1, k) 1, and if h (1, k) is smaller than or equal to 0, outputting h (1, k) 0
(step 5) k is k +1, and whether k is less than or equal to the maximum test step number is judged, if yes, the step 1 is returned to
Wherein
The mean absolute error of the model training calculated in step 2. Therefore, the prediction output of the prediction model in the next step is realized by using the prediction error in the previous step.
Compared with the prior art, the invention has the beneficial effects that:
1. the denoising processing based on the wavelet algorithm is carried out by setting a double-layer threshold, the singular point coefficient is firstly shrunk for a plurality of times to return to the normal range, and then the global signal is subjected to soft threshold processing, so that the singular point problem can be solved well. And the traditional soft threshold method only shrinks the coefficient more than the threshold value once, so that the effect of removing singular points is not ideal.
2. The oxygen content of the flue gas which is difficult to directly and accurately measure is predicted by using a soft measurement method, frequent replacement of a zirconia sensor is avoided, and the measurement cost is saved.
3. The prediction model is established by using the dynamic neural network Elman (PSO-Elman model) optimized by the particle swarm optimization, so that the nonlinear model of boiler combustion can be better approximated, the local minimum can be avoided in the iterative operation of the algorithm, and the convergence speed of the iterative operation is accelerated.
4. The training set data used for model training is selected from the running conditions under various working conditions, so that the trained prediction model is suitable for various working conditions and has better generalization capability.
5. An error correction feedback link is added, so that the prediction model can be adjusted on line, model mismatch caused by disturbance and other conditions in model online test is avoided, and accuracy and stability of the prediction model are guaranteed.
Drawings
FIG. 1 is a schematic diagram of the PSO-Elman offline prediction model algorithm of the present invention;
FIG. 2 is a schematic diagram of the feedback corrected PSO-Elman online prediction model algorithm of the present invention;
FIG. 3 is a diagram of the PSO-Elman model iterative error change process of the present invention;
FIG. 4 is a comparison graph of the online test effects of the feedback correction PSO-Elman model and the PSO-Elman model of the present invention;
FIG. 5 is a graph of the feedback corrected PSO-Elman model on-line test prediction error of the present invention;
FIG. 6 is a graph of the feedback corrected PSO-Elman model on-line test predicted relative error of the present invention;
FIG. 7 is a block diagram of the overall system of the present invention;
FIG. 8 is a schematic diagram of the upper computer operation of the present invention;
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail with reference to embodiments, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. 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 invention.
The present invention will be described in detail with reference to specific embodiments. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of protection of the present invention
As shown in the figures 1 and 2, the invention provides an online prediction method for the oxygen content of flue gas of a thermal power plant based on a PSO-Elman model. The method mainly adopts the idea that a PSO-Elman model is utilized to approximate a complex boiler combustion nonlinear model and error feedback correction so as to realize online correction of a prediction model to achieve an online accurate prediction function, and mainly comprises two parts: and (3) establishing an offline prediction model and performing an online model correction algorithm. The off-line prediction model establishment comprises the following three steps: reading historical data from a DCS (distributed control system) of a thermal power plant, selecting relevant auxiliary variables and target prediction variables for modeling, dividing the auxiliary variables and the target prediction variables into a training set and an online test set, and normalizing the data; then establishing a prediction model between related auxiliary variables and the oxygen content of the smoke by using an Elman neural network, and setting the fitness of a particle swarm optimization algorithm; and finally, introducing a PSO optimization algorithm to optimize all weights and thresholds in the Elman network structure to obtain optimal weights and thresholds, and endowing the optimal weights and thresholds to the Elman model to complete the construction of the PSO-Elman model. The online model correction algorithm is to use an online test set to operate a PSO-Elman flue gas oxygen content prediction model, calculate the prediction error of each step, and perform weighted feedback on the prediction error of each step to the next prediction output to complete the feedback correction of the online prediction output.
The method comprises the following specific steps:
(1) firstly, SG-2136.5/17.55-M type 660MW subcritical boiler combustion data are read from a DCS (distributed control System) of a certain thermal power plant in Jiaxing, Zhejiang, and correlation coefficients of variables and outlet oxygen quantity (namely, flue gas oxygen content) of an air preheater are calculated by using a Pearson correlation coefficient method, and when the correlation coefficients are higher than 8.0, the correlation coefficients are considered to have high correlation with the flue gas oxygen content. The calculation formula is as follows:
(2) method for denoising main steam pressure, main steam flow and total wind by using improved wavelet double-layer thresholdAnd noise and singular points are removed from a data set consisting of the quantity, the outlet flue gas temperature of the air preheater, the total coal quantity, the boiler load, the current of the air feeder and the oxygen content of the flue gas. Based on the traditional noise estimation and the general threshold calculation, the first threshold T is larger than the first threshold1Wavelet coefficient contraction bit of1In, i.e.
|dj(n)|＞T1Then re-estimating the noise formation T2By means of T2And carrying out soft threshold processing on the signal in the last step, wherein a value taking function is as follows:
wherein
The wavelet coefficients after two threshold adjustments are respectively.
(3) And selecting training set data by a hierarchical sampling method. Because the mapping relation between the input data and the oxygen content of the smoke in the daily electricity utilization peak period is more complex, the daily data are divided into two layers (namely 7: 00-23: 00; 00: 00-07: 00 and 23: 00-24: 00) according to the daily electricity utilization peak period of 7: 00-23: 00. Since the system sampling frequency is 15S/time, and the sample size of the first layer data (7: 00-23: 00) is 3840, calculating the sampling ratio of the layer:
the sample size of the second layer data (00: 00-07: 00 and 23: 00-24: 00) is 1920, then the layer sampling ratio is calculated as:
thus, it is determined that the sampling ratios of the two layers of data are set to 2/3 and 1/3, respectively, 900 sets of data are extracted to form a training set for training the model. And a continuous 800 sets of data were selected for the test set for online verification.
(4) Establishing an Elman structure: the input is 7 parameters, the output is 1 parameter, the Elman network structure is established to be 7-15-1, the hidden layer transfer function uses a tansig function, the output layer transfer function uses a linear function, the training function is tranlm, namely, the formula for correcting the network weight by adopting a Levenberg-Marquardt (L-M) algorithm is as follows:
ω(n+1)＝ω(n)-[JTJ+μI]-1JTe
wherein e is an error performance function, and J is a Jacobian matrix of the first derivative of the error performance function to the network weight. By obtaining the error of the training, the average absolute error is calculated
Where N is the number of training sessions.
(5) Optimizing all weights and thresholds in the Elman network structure by introducing a PSO optimization algorithm, firstly defining a prediction error as a fitness value in the PSO algorithm, and calculating the formula as follows:
and performing population random initialization, wherein the population size of the particle group is set to be 40, the particle length is set to be 105, the maximum iteration number is set to be 100, the learning factor C1 is 2.3, the C2 is 1.3, the maximum speed vmax of the particle motion is 1, the maximum inertia weight wmax is 0.9, and the minimum wmin is 0.3. Then calculating the fitness of the individual, and selecting the local optimal position vector of the individual and the global optimal position vector of the population; then, the iteration times, each individual velocity vector and each individual position vector are set; then for each individual speed update: vid＝ωVid+C1r1(Pid-Xid)+C2r2(Pgd-Xid) And displacement updating: xid＝Xid+VidWherein Vid is the example movement speed; xid is the current position of the particle; omega is an inertia factor; c1 and C2 are learning factors, and C1 is generally equal to C2 is equal to 2; r1, r2 is [0, 1]]A random number of (c); pid is the individual extremum for the ith example; pgd is a global extremum. And finally, judging whether the maximum iteration times or the global optimal position is reached, outputting the optimal solution after the maximum iteration times or the global optimal position is reached, and returning to continue the iteration.And obtaining the optimal weight and the threshold, and endowing the optimal weight and the threshold to the Elman model to complete the construction of the PSO-Elman model.
(6) Setting the test time length to be 175 minutes (sampling frequency is 15 seconds/time), gradually reading data from the online test set, performing online test on the PSO-Elman prediction model, and utilizing the prediction error e (k-1) of the k-1 second, namely
Weighted correction kth prediction: setting a weighting coefficient matrix h with dimension of 1 × 800, wherein a formula corresponding to the k-th second prediction output correction is as follows:
therefore, the prediction output of the next prediction model is corrected by using the prediction error of the previous step. Setting the initial values of all elements in the weighting coefficient matrix h to be 0.5, and the value interval to be [0.1], and then the pseudo code of the weighting coefficient updating algorithm is as follows:
wherein
The mean absolute error of the model training calculated in step 2. Therefore, the prediction output of the next prediction model is corrected by using the prediction error of the previous step.
FIG. 3 shows the iterative error change rate in the PSO algorithm evolution process, the iterative error of the PSO-Elman model is in the order of-2, and the model has very high prediction accuracy. After the 20 th iteration, the iteration error rate of change drops below 0.04 and quickly settles to around 0.035.
Fig. 4 is a comparison of the online test effects of the feedback correction PSO-Elman model and the PSO-Elman model, and it can be seen that the feedback correction PSO-Elman model can more closely track the real flue gas oxygen content curve and has smaller fluctuation as the real flue gas oxygen content value changes.
The error and the relative error of the feedback correction PSO-Elman model online test are shown in FIGS. 5 and 6 respectively, and can be seen to be less than +/-0.05, and the relative error range is within +/-0.15.
Fig. 7 and 8 show a system for predicting the oxygen content in flue gas of a thermal power plant, which is provided by the invention, and the system comprises a boiler of the thermal power plant, a distributed control system for controlling the operation of the boiler, a data communication interface, a database and an upper computer. The database reads data from the distributed control system through the data communication interface and is used for training, learning and testing of the upper computer, the upper computer exchanges data with the distributed control system through the data communication interface, and the upper computer comprises an offline learning training part, an online prediction verification part, an error feedback correction part and a model updating judgment part. Wherein the host computer body includes:
(1) the signal acquisition module: the module is used for collecting boiler operation state parameters from a boiler distributed control system, and selecting main steam pressure, main steam flow, total air quantity, air preheater outlet flue gas temperature, total coal quantity, boiler load and blower current phase as input variables of a flue gas oxygen content prediction model by calculating Pearson correlation coefficients.
(2) A data preprocessing module: the module removes singular points and noise from the input data of the prediction model by using an improved wavelet double-layer threshold denoising method. And establishing training set data and on-line test set data through hierarchical sampling, and then carrying out normalization processing on the data, and mapping the data into the range of [ -1,1 ].
(3) An expert knowledge base: the module continuously updates the training samples by adopting a method of rolling a time window, so that the training samples are always kept in the latest state, wherein the method of rolling the time window refers to the step of backtracking the time size of L length from the current time, and the unit of L is second;
(4) an intelligent modeling module: establishing a relation model between related auxiliary variables and the oxygen content of the smoke by using an Elman neural network, introducing a PSO algorithm to optimize the weight and the threshold in the Elman model to obtain an optimal weight and an optimal threshold, endowing the optimal weight and the optimal threshold to the Elman network model, and completing construction of a PSO-Elman prediction model, wherein the specific process comprises the following steps:
4.1) Elman structure was established and trained: the input is 7 parameters, the output is 1 parameter, the Elman network structure is established to be 7-15-1, the hidden layer transfer function uses a tansig function, the output layer transfer function uses a linear function, the training function is tranlm, namely, the formula for correcting the network weight by adopting a Levenberg-Marquardt (L-M) algorithm is as follows:
ω(n+1)＝ω(n)-[JTJ+μI]-1JTe
wherein e is an error performance function, and J is a Jacobian matrix of the first derivative of the error performance function to the network weight. By obtaining the error of the training, the average absolute error is calculated
Where N is the number of training sessions.
4.2) introducing a PSO optimization algorithm to optimize all weights and thresholds in the Elman network structure, firstly defining a prediction error as a fitness value in the PSO algorithm, and calculating the formula as follows:
and performing population random initialization, wherein the population size of the particle group is set to be 40, the particle length is set to be 105, the maximum iteration number is set to be 100, the learning factor C1 is 2.3, the C2 is 1.3, the maximum speed vmax of the particle motion is 1, the maximum inertia weight wmax is 0.9, and the minimum wmin is 0.3. Then calculating the fitness of the individual, and selecting the local optimal position vector of the individual and the global optimal position vector of the population; then, the iteration times, each individual velocity vector and each individual position vector are set; then for each individual speed update: vid＝ωVid+C1r1(Pid-Xid)+C2r2(Pgd-Xid) And displacement updating: xid＝Xid+VidWherein Vid is the example movement speed; xid is the current position of the particle; omega is an inertia factor; c1 and C2 are the factors of learningC1 ═ C2 ═ 2; r1, r2 is [0, 1]]A random number of (c); pid is the individual extremum for the ith example; pgd is a global extremum. And finally, judging whether the maximum iteration times or the global optimal position is reached, outputting the optimal solution after the maximum iteration times or the global optimal position is reached, and returning to continue the iteration. And obtaining the optimal weight and the threshold, and endowing the optimal weight and the threshold to the Elman model to complete the construction of the PSO-Elman model.
(5) An error correction feedback module: and operating a PSO-Elman flue gas oxygen content prediction model by using the online test set, calculating a prediction error of each step, and performing weighted feedback on the prediction error of each step to the next prediction output to finish feedback correction on the online prediction output. The specific method comprises the following steps:
using the prediction error e (k-1) of the k-1 second, i.e.
Weighted correction kth prediction: setting a weighting coefficient matrix h with dimension of 1 × 800, wherein a formula corresponding to the k-th second prediction output correction is as follows:
setting the initial values of all elements in the weighting coefficient matrix h to be 0.5, and the value interval to be [0.1], and then the pseudo code of the weighting coefficient updating algorithm is as follows:
wherein
The mean absolute error of the model training calculated in step 2. Therefore, the prediction output of the next prediction model is corrected by using the prediction error of the previous step.
(6) A model evaluation module: and obtaining a prediction error of each step of the model, judging whether the error is more than or equal to 5%, if so, considering that the current prediction model is mismatched, returning to the intelligent modeling module, and training the model again by using the latest training set data from the expert knowledge base.
(7) A communication module: the module transmits the prediction result of the oxygen content of the flue gas meeting the precision requirement to a boiler distributed control system.
The present invention has been described in terms of specific examples, which are provided to aid understanding of the invention and are not intended to be limiting. Any partial modification or replacement within the technical scope of the present disclosure by a person skilled in the art should be included in the scope of the present disclosure.
Claims (2)
1. Thermal power plant's flue gas oxygen content online prediction system, its characterized in that: the system comprises an upper computer and a data communication interface, wherein the data communication interface is connected with a boiler distributed control system; establishing and updating a flue gas oxygen content prediction model in the upper computer, and transmitting a prediction result meeting the precision requirement to a distributed control system through a data communication interface;
the host computer includes:
the signal acquisition module: the module is used for collecting boiler operation state parameters from a boiler distributed control system, and selecting main steam pressure, main steam flow, total air volume, air preheater outlet flue gas temperature, total coal volume, boiler load and blower current phases as input variables of a flue gas oxygen content prediction model by calculating Pearson correlation coefficients;
a data preprocessing module: the module removes singular points and noise from input data of a prediction model by using an improved wavelet double-layer threshold denoising method; establishing training set data and on-line test set data through layered sampling, and then carrying out normalization processing on the data, and mapping the data into a range of [ -1,1 ];
an expert knowledge base: the module continuously updates the training samples by adopting a method of rolling a time window, so that the training samples are always kept in the latest state, wherein the method of rolling the time window refers to the step of backtracking the time size of L length from the current time, and the unit of L is second;
an intelligent modeling module: the model establishes a relation model between related auxiliary variables and the oxygen content of smoke by using an Elman neural network, then introduces a PSO algorithm to optimize the weight and the threshold in the Elman model to obtain the optimal weight and the optimal threshold, and gives the optimal weight and the optimal threshold to the Elman network model to complete the construction of a PSO-Elman prediction model;
an error correction feedback module: the module utilizes an online test set to operate a PSO-Elman flue gas oxygen content prediction model, calculates prediction errors of each step, and feeds the prediction errors of each step back to next prediction output in a weighted mode to finish feedback correction of the online prediction output;
a model evaluation module: the module obtains a prediction error of each step of the model, judges whether the error is more than or equal to 5 percent, if so, considers that the current prediction model is mismatched, returns to the intelligent modeling module, and trains the model again by using the latest training set data from the expert knowledge base;
a communication module: the module transmits the prediction result of the oxygen content of the flue gas meeting the precision requirement to a boiler distributed control system.
2. The prediction method of the prediction system according to claim 1, characterized in that: the method realizes online prediction of the oxygen content of the flue gas by establishing a PSO-Elman prediction model and introducing a feedback correction structure, and comprises the following steps:
step 1: reading historical data from a distributed control system of a thermal power plant, selecting relevant auxiliary variables and target predictive variables for modeling, dividing the auxiliary variables and the target predictive variables into a training set and an online test set, and normalizing the data;
step 2: establishing a prediction model between related auxiliary variables and the oxygen content of the smoke by using an Elman neural network, and training to further calculate a prediction average absolute error;
and step 3: optimizing all weights and thresholds in the Elman network structure by introducing a PSO optimization algorithm to obtain optimal weights and thresholds, and endowing the optimal weights and thresholds to the Elman model to complete the construction of the PSO-Elman model;
and 4, step 4: and operating a PSO-Elman flue gas oxygen content prediction model by using the online test set, calculating a prediction error of each step, and performing weighted feedback on the prediction error of each step to the next prediction output to finish feedback correction on the online prediction output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011179665.5A CN112308311A (en) | 2020-10-29 | 2020-10-29 | Online prediction system and method for oxygen content of flue gas of thermal power plant |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011179665.5A CN112308311A (en) | 2020-10-29 | 2020-10-29 | Online prediction system and method for oxygen content of flue gas of thermal power plant |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112308311A true CN112308311A (en) | 2021-02-02 |
Family
ID=74330405
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011179665.5A Pending CN112308311A (en) | 2020-10-29 | 2020-10-29 | Online prediction system and method for oxygen content of flue gas of thermal power plant |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112308311A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114202065A (en) * | 2022-02-17 | 2022-03-18 | 之江实验室 | Stream data prediction method and device based on incremental evolution LSTM |
| CN116822712A (en) * | 2023-05-25 | 2023-09-29 | 华能国际电力股份有限公司上海石洞口第二电厂 | CVaR-based thermal power plant fire coal purchasing optimization method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100154389A1 (en) * | 2008-02-08 | 2010-06-24 | Schumacher Sascha | Procedure for regenerating an exhaust gas after treatment system |
| CN106931453A (en) * | 2017-02-27 | 2017-07-07 | 浙江大学 | The forecasting system and method for circulating fluid bed domestic garbage burning emission of NOx of boiler |
| CN107016455A (en) * | 2017-02-27 | 2017-08-04 | 浙江大学 | The forecasting system and method for circulating fluid bed domestic garbage burning boiler furnace outlet flue gas oxygen content |
| CN109492319A (en) * | 2018-11-23 | 2019-03-19 | 东北电力大学 | A kind of power plant boiler flue gas oxygen content flexible measurement method |
-
2020
- 2020-10-29 CN CN202011179665.5A patent/CN112308311A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100154389A1 (en) * | 2008-02-08 | 2010-06-24 | Schumacher Sascha | Procedure for regenerating an exhaust gas after treatment system |
| CN106931453A (en) * | 2017-02-27 | 2017-07-07 | 浙江大学 | The forecasting system and method for circulating fluid bed domestic garbage burning emission of NOx of boiler |
| CN107016455A (en) * | 2017-02-27 | 2017-08-04 | 浙江大学 | The forecasting system and method for circulating fluid bed domestic garbage burning boiler furnace outlet flue gas oxygen content |
| CN109492319A (en) * | 2018-11-23 | 2019-03-19 | 东北电力大学 | A kind of power plant boiler flue gas oxygen content flexible measurement method |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114202065A (en) * | 2022-02-17 | 2022-03-18 | 之江实验室 | Stream data prediction method and device based on incremental evolution LSTM |
| CN116822712A (en) * | 2023-05-25 | 2023-09-29 | 华能国际电力股份有限公司上海石洞口第二电厂 | CVaR-based thermal power plant fire coal purchasing optimization method and system |
| CN116822712B (en) * | 2023-05-25 | 2024-04-09 | 华能国际电力股份有限公司上海石洞口第二电厂 | CVaR-based thermal power plant fire coal purchasing optimization method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102778538B (en) | Soft measuring method based on improved SVM (Support Vector Machine) for measuring boiler unburned carbon content in fly ash | |
| CN112308311A (en) | Online prediction system and method for oxygen content of flue gas of thermal power plant | |
| CN112580250A (en) | Thermal power generating unit denitration system based on deep learning and optimization control method | |
| CN113433911B (en) | Accurate control system and method for ammonia spraying of denitration device based on accurate concentration prediction | |
| CN112149879B (en) | New energy medium-and-long-term electric quantity prediction method considering macroscopic volatility classification | |
| CN100545772C (en) | A kind of coal-burning boiler system mixing control method | |
| KR20130099479A (en) | Method of sensorless mppt neural control for wind energy conversion systems | |
| CN112283689A (en) | On-line monitoring system and detection method for accumulated ash on heating surface of coal-fired power station boiler | |
| CN110207094A (en) | IQGA-SVR boiler heating surface fouling characteristics discrimination method based on principal component analysis | |
| CN108227759A (en) | A kind of solar energy tracking control system and method based on neural network technology | |
| CN112016754A (en) | Power station boiler exhaust gas temperature advanced prediction system and method based on neural network | |
| CN114721263B (en) | Intelligent regulation and control method for cement decomposing furnace based on machine learning and intelligent optimization algorithm | |
| CN115510904A (en) | Boiler heating surface ash deposition monitoring method based on time sequence prediction | |
| CN108762086B (en) | Secondary reheat steam temperature control device and control system based on model predictive control | |
| CN114139439A (en) | Steam turbine optimal initial pressure determination method based on simulated annealing particle swarm algorithm | |
| CN116865343A (en) | Model-free self-adaptive control method, device and medium for distributed photovoltaic power distribution network | |
| CN115656439A (en) | Online monitoring method for nitrogen oxide emission concentration of coal-fired unit based on transfer learning | |
| CN117270387A (en) | SCR denitration system low ammonia escape control method and system based on deep learning | |
| Yang et al. | Application of fuzzy neural network PID algorithm in oil pump control | |
| CN112947606A (en) | Boiler liquid level control system and method based on BP neural network PID predictive control | |
| CN114462688A (en) | Tube explosion detection method based on LSTM model and dynamic threshold determination algorithm | |
| CN114397813A (en) | Power generation boiler combustion continuous sliding film control method based on slow time-varying disturbance observer | |
| CN111695082A (en) | Anti-differential state estimation method for intelligent power distribution network | |
| CN111538355B (en) | GA-IGPC-based boiler flue GAs oxygen content control method and system | |
| CN110794679A (en) | Prediction control method and system for load regulation of industrial steam supply system |
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 |