CN103870878B - A kind of Combustion Characteristics in Utility Boiler neural network model - Google Patents
A kind of Combustion Characteristics in Utility Boiler neural network model Download PDFInfo
- Publication number
- CN103870878B CN103870878B CN201410124690.1A CN201410124690A CN103870878B CN 103870878 B CN103870878 B CN 103870878B CN 201410124690 A CN201410124690 A CN 201410124690A CN 103870878 B CN103870878 B CN 103870878B
- Authority
- CN
- China
- Prior art keywords
- burner
- input
- burners
- sets
- neuron
- 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
Abstract
A kind of Combustion Characteristics in Utility Boiler neural network model, including mode input, model output, neutral net；Mode input includes each burner output, the ature of coal parameter of burner correspondence coal dust, coal powder density, flow velocity, First air, secondary wind, burnout degree, oxygen amount, load, carbon monoxide content, combustion chamber draft, each the pressure of bellows and wind powder flow path each governor motion aperture；Model output includes that the stove effect weighing boiler performance or stove relatively are imitated and characterize the NOx content of discharge；Mode input is divided into the exclusive input of burner according to physical characteristic and other global impact inputs, corresponding different types of input layer and hidden layer neuron；Station boiler thaumatropy can be corresponding neutral net special construction by the present invention, network model is made to comprise the information of actual boiler physical characteristic, avoid unnecessary mutual and coupling, the sub-network parameter of similar characteristics can be made to reuse simultaneously, there is higher performance and efficiency.
Description
Technical field
The present invention relates to Combustion Characteristics in Utility Boiler modeling technique field, be specifically related to a kind of power boiler burning
Characteristic neural network model.
Background technology
Boiler combustion system is the most complicated part being most difficult to modeling of electric power station system, and it relates to answering in large space field
Miscellaneous physical and chemical changes, the transmission of its heat causes again the phase transformation of working medium, and the high temperature two phase flow of burning lacks
Weary effective means of testing, therefore, it is difficult to be been described by by classical mathematics method, there is no the warp of mature and reliable
Allusion quotation mathematical model.Nerual network technique is suitable for Complex Nonlinear System and self-learning property in a large amount of works with it
The modeling field of industry system achieves and is widely applied.
The more modeling concentrating on soda pop side of the mathematical model of boiler, but this class model is difficult to reflect boiler
Efficiency of combustion, puts into and the impact of air distribution distribution to describe the coal dust of coal-burning boiler with mathematical model, it is necessary to
Set up the mathematical model evaluating boiler combustion characteristic, reflection boiler air, powder system input and boiler combustion efficiency
Relation between output.
The prior art overwhelming majority uses feedforward multilayer neural network or neutral net sequence to simulate boiler
Combustibility, the hidden layer neuron quantity that the possible employing of different modeled example is different, but this kind of nerve net
Network belongs to classical neural network structure, and has the self study of maturation and training algorithm available, greatly
Amount industrial process modeling obtains application, but there is following problem for the model of boiler combustion performance:
1. boiler combustion input parameter many, system complex, need relatively multi-neuron network consisting, its training and
Study is required for the sample of magnanimity, and the boiler running process operating mode of reality usually fluctuates, it is difficult to
Sufficient amount of sample is obtained in short time；
2. conventional neural networks Model suitability is extensive, but cannot the special construction of steam generator system and physics be advised
Rule is converted into neural network structure or parameter, and hidden layer neuron quantity is difficult to determine, modeling process
Complicated and time consumption, initial weight matrix value effect is big, is trained to power low；
3. steam generator system input quantity is many, but dependency between different input or very strong or the most weak, right
Symmetrical structure (output of preceding layer neuron and next layer used in conventional neural networks technology
Each neuron have connection) the easily failure because of the precocity of study for the training of model.
Summary of the invention
In order to solve the problem that above-mentioned prior art exists, it is special that the present invention proposes a kind of new power boiler burning
Property neural network model, can be corresponding neutral net special construction by station boiler thaumatropy, it is to avoid no
Necessary mutual and coupling, makes the sub-network parameter of similar characteristics to reuse, not only introduces boiler physics
The information of principle and model dependence to sample is greatly reduced, be remarkably improved neural network model modeling,
Training, the learning efficiency and model are applied to power.
To achieve these goals, the present invention is by the following technical solutions:
A kind of Combustion Characteristics in Utility Boiler neural network model, including mode input, model output and nerve net
Network；Described mode input includes that each burner output, the ature of coal parameter of burner correspondence coal dust, coal dust are dense
Degree, flow velocity, First air, secondary wind, burnout degree, oxygen amount, load, carbon monoxide content, combustion chamber draft,
Each the pressure of bellows and wind powder flow path each governor motion aperture；The output of described model selects to weigh the stove of boiler performance
Effect or relatively stove effect and the NOx content of sign discharge；Described neutral net uses sets of burners independent of each other
The neural tuple of son each sets of burners corresponding, the exclusive input layer in each sets of burners only with belong to
The hidden layer neuron of this group is connected；Each sets of burners has the burning of analogous location, structure or physical characteristic
Device group forces to use identical network structure and weighting parameter, is embodied as network structure and includes following three classes:
The first kind: hidden layer neuron is divided into N group burner and other system N+1 part, N number of combustion altogether
The exclusive input layer of burner group is only connected with the hidden layer neuron belonging to this group burner, burner
Between group separate, with global impact sexual system input the most separate, all hidden layer neuron export
It is connected two-by-two with output layer neuron；
Equations of The Second Kind: hidden layer neuron is divided into N group burner and other system N+1 part, N number of combustion altogether
The exclusive input layer of burner group is only connected with the hidden layer neuron belonging to this group burner, burner
Between group separate, global impact sexual system input be then connected two-by-two with each hidden layer neuron, institute
Hidden layer neuron output is had to be connected two-by-two with output layer neuron；
3rd class: network structure comprises two hidden layers, the neuron of the first hidden layer is divided into the combustion of N group
Burner N section altogether, the exclusive input layer of N number of sets of burners is the most hidden with belong to this group burner first
Neuron containing layer is connected, separate between sets of burners, the neuron of the first hidden layer and global impact
Sexual system input neuron is connected two-by-two with each second hidden layer neuron, and all second hidden layers are neural
Unit's output is connected two-by-two with output layer neuron.
All mode inputs are divided into two classes, the ature of coal parameter of each sets of burners according to physical principle by described model
(including caloric value, ash, moisture, volatile matter, each constituent content) and powder concentration, primary air velocity, wind
Powder mixture temperature, secondary air flow, secondary air damper aperture have stronger with the burning of corresponding sets of burners
Impact, and more weak with the combustion effects of other sets of burners, as the input of this sets of burners exclusive；After-flame
The burning of each sets of burners is had certain by wind, oxygen amount, load, carbon monoxide content, combustion chamber draft
Impact, inputs as global impact.Two kinds of input signals are distinguished in described model and come.
According to the station boiler type of furnace and the difference of control condition, actually enter a subset of preference pattern input；
One subset of actual output preference pattern output；The input of actual exclusive burner selects above-mentioned exclusive burning
One subset of device group input signal.
The special neutral net of above structure neutral net compared to existing technology is used to have the advantage that
First this neutral net can natural differentiation difference input parameter classification, the exclusive input of sets of burners is right
The impact of other burner is the least, and the burning of whole boiler is all had by the global impact parameters such as above-mentioned oxygen amount
Having large effect, the input of this two class has natural weight difference in model structure, will affect whole pot
The signal inputs exclusive with the burner affecting partial combustion such as the oxygen amount of stove burning distinguish, and make neutral net exist
With regard to the information of the natural exclusive construction features of succession boiler before study and training；Secondly the division of sets of burners is used,
The sets of burners physically with similar characteristic can be used identical local network structure and weights, the most right
In the bottom front-back wall sets of burners of opposed firing boiler it is believed that its exclusive input is to whole boiler combustion
Impact is similar, and its network model should have similar structure and parameter, network so can be greatly reduced and learning
Practise and during training, need the number of parameters solved, and avoiding similar physical structure to draw the parameter that differs greatly
The unreasonable result of result, so that the training speed of network model and success rate are increased dramatically；Finally
Due to the classical neural network that uses compared to existing technology, (output of every layer of neuron is with next layer of neuron two-by-two
Interconnection) parameter that solves in the training process is greatly reduced, is also greatly reduced the demand of required sample size,
The training of ad eundem can be realized through analyzing the sample averagely only needing identical scale classic network 1/8 the most less
Effect.
Due to the particularity of this Artificial Neural Network Structures, therefore conventional classic network study and training algorithm
Cannot apply, the neutral net intelligence similar based on " genetic algorithm " or " particle cluster algorithm " can only be selected
Study and training algorithm, this kind of intelligent algorithm has achieved a large amount of in the study of classical neural network and training
Application, various improved algorithm emerge in an endless stream, and all can be applicable to neural network structure proposed by the invention
Study and training, although the speed of its study and training is slightly lower compared with classic algorithm, but its robustness is good, can lead to
The control crossing search volume is avoided the precocity of training and dissipates, and is more suitable for industrial automation than classic algorithm
The application of modeling.
Accompanying drawing explanation
Fig. 1 is that the first kind implements network structure.
Fig. 2 is that Equations of The Second Kind implements network structure.
Fig. 3 is that the 3rd class implements network structure.
Detailed description of the invention
The present invention will be described in more detail with specific embodiment below in conjunction with the accompanying drawings.
One Combustion Characteristics in Utility Boiler neural network model of the present invention, including mode input, model output,
Neutral net；The input of described neural network model includes each burner output, the ature of coal of burner correspondence coal dust
Parameter, coal powder density, flow velocity, First air, secondary wind, burnout degree, oxygen amount, load, carbon monoxide contain
Amount, combustion chamber draft, each the pressure of bellows and wind powder flow path each governor motion aperture；The output of described model is optional
Stove effect or the stove relatively of weighing boiler performance are imitated and characterize the NOx content of discharge.
All mode inputs are divided into two classes, the ature of coal parameter of each sets of burners according to physical principle by described model
(including caloric value, ash, moisture, volatile matter, each constituent content) and powder concentration, primary air velocity, wind
Powder mixture temperature, secondary air flow, secondary air damper aperture have stronger with the burning of corresponding sets of burners
Impact, and more weak with the combustion effects of other sets of burners, as the input of this sets of burners exclusive；After-flame
The burning of each sets of burners is had certain by wind, oxygen amount, load, carbon monoxide content, combustion chamber draft
Impact, inputs as global impact；Two kinds of input signals are distinguished in described model and come.
Described neutral net uses sets of burners sub-nerve tuple independent of each other each sets of burners corresponding,
Exclusive input layer in each sets of burners is only connected with the hidden layer neuron belonging to this group；Each burning
Device group has the sets of burners of analogous location, structure or physical characteristic forces to use identical network structure and power
Value parameter, is embodied as network structure and includes following three classes.
Be illustrated in figure 1 the first kind implement network structure: hidden layer neuron be divided into N group burner with
Other system altogether N+1 part, the exclusive input layer of N number of sets of burners only with belong to this group burner
Hidden layer neuron is connected, separate between sets of burners, with the input of global impact sexual system also the most solely
Vertical, the output of all hidden layer neuron is connected two-by-two with output layer neuron；
As shown in Figure 2 Equations of The Second Kind implement network structure: hidden layer neuron be divided into N group burner and its
His system N+1 part altogether, the exclusive input layer of N number of sets of burners only with belong to the hidden of this group burner
Being connected containing layer neuron, separate between sets of burners, the input of global impact sexual system is then hidden with each
The neuron containing layer is connected two-by-two, and the output of all hidden layer neuron is connected two-by-two with output layer neuron；
3rd class implements network structure as shown in Figure 3: network structure comprises two hidden layers, the first hidden layer
Neuron be divided into N group burner altogether N section, the exclusive input layer of N number of sets of burners only with
The neuron of the first hidden layer belonging to this group burner is connected, and separate between sets of burners, first is hidden
Neuron and global impact sexual system input neuron containing layer are biphase with each second hidden layer neuron two
Even, all second hidden layer neuron outputs are connected two-by-two with output layer neuron.
As the preferred embodiment of the present invention, according to the station boiler type of furnace and the difference of control condition, actual
One subset of the optional above input of input；One subset of the optional above-mentioned output of actual output；Actual
One subset of the optional above-mentioned exclusive burner input signal of input of exclusive burner.
Embodiment
Certain high-power plant boiler is opposed firing pattern, front-back wall each three rows totally six groups of burners, layering
Coal mixing combustion, every grate firing burner is furnished with a secondary air box, and two ends arrange secondary air damper, and pulverized coal preparation system revolves
Wind skimming baffle can not adjust automatically, and swirl secondary air pull bar is manual.According to above boiler type, structure
Feature, the condition automatically controlled and modeling target, select a subset of input signal described in the inventive method
(load, oxygen amount, sets of burners primary air flow, each layer secondary air damper aperture, each layer ature of coal, respectively burn
Device exerts oneself, front and back burnout degree baffle opening) as mode input, the most each layer ature of coal, burner output,
Layer secondary air damper aperture exports as model as the exclusive input of sets of burners, boiler efficiency, uses such as Fig. 1
Shown network structure, employing classical particle group's algorithm is as e-learning and training algorithm, according to boiler operatiopn
During obtain sample be trained, be i.e. available for boiler combustion efficiency analyze neural network model.
Claims (2)
1. a Combustion Characteristics in Utility Boiler neural network model, it is characterised in that: include mode input, mould
Type output and neutral net；Described mode input includes each burner output, the coal of burner correspondence coal dust
Matter parameter, coal powder density, flow velocity, First air, secondary wind, burnout degree, oxygen amount, load, carbon monoxide
Content, combustion chamber draft, each the pressure of bellows and wind powder flow path each governor motion aperture；The output of described model selects
Stove effect or the stove relatively of weighing boiler performance are imitated and characterize the NOx content of discharge；Described neutral net uses that
This independent sets of burners sub-nerve tuple each sets of burners corresponding, the exclusive input in each sets of burners
Layer neuron is only connected with the hidden layer neuron belonging to this group；Each sets of burners has analogous location, structure
Or the sets of burners pressure of physical characteristic uses identical network structure and weighting parameter, it is embodied as network knot
Structure includes following three classes:
The first kind: hidden layer neuron is divided into N group burner and other system N+1 part, N number of combustion altogether
The exclusive input layer of burner group is only connected with the hidden layer neuron belonging to this group burner, burner
Between group separate, with global impact sexual system input the most separate, all hidden layer neuron export
It is connected two-by-two with output layer neuron；
Equations of The Second Kind: hidden layer neuron is divided into N group burner and other system N+1 part, N number of combustion altogether
The exclusive input layer of burner group is only connected with the hidden layer neuron belonging to this group burner, burner
Between group separate, global impact sexual system input be then connected two-by-two with each hidden layer neuron, institute
Hidden layer neuron output is had to be connected two-by-two with output layer neuron；
3rd class: network structure comprises two hidden layers, the neuron of the first hidden layer is divided into the combustion of N group
Burner N section altogether, the exclusive input layer of N number of sets of burners is the most hidden with belong to this group burner first
Neuron containing layer is connected, separate between sets of burners, the neuron of the first hidden layer and global impact
Sexual system input neuron is connected two-by-two with each second hidden layer neuron, and all second hidden layers are neural
Unit's output is connected two-by-two with output layer neuron；
It is divided into two classes, the ature of coal parameter of each sets of burners to include heating all mode inputs according to physical principle
Amount, ash, moisture, volatile matter and each constituent content and powder concentration, primary air velocity, wind powder mixture temperature
Degree, secondary air flow, secondary air damper aperture have a stronger impact with the burning of corresponding sets of burners, and with
The combustion effects of other sets of burners is more weak, as the input of this sets of burners exclusive；Burnout degree, oxygen amount,
The burning of each sets of burners is had a certain impact by load, carbon monoxide content, combustion chamber draft, as
Global impact inputs；Two kinds of input signals are distinguished in described model and come.
A kind of Combustion Characteristics in Utility Boiler neural network model the most according to claim 1, its feature exists
In: according to the station boiler type of furnace and the difference of control condition, actually enter a subset of preference pattern input；
One subset of actual output preference pattern output；The input of actual exclusive burner selects described exclusive burning
One subset of device group input signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410124690.1A CN103870878B (en) | 2014-03-28 | 2014-03-28 | A kind of Combustion Characteristics in Utility Boiler neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410124690.1A CN103870878B (en) | 2014-03-28 | 2014-03-28 | A kind of Combustion Characteristics in Utility Boiler neural network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103870878A CN103870878A (en) | 2014-06-18 |
| CN103870878B true CN103870878B (en) | 2016-09-28 |
Family
ID=50909391
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410124690.1A Active CN103870878B (en) | 2014-03-28 | 2014-03-28 | A kind of Combustion Characteristics in Utility Boiler neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103870878B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10679119B2 (en) * | 2017-03-24 | 2020-06-09 | Intel Corporation | Handling signal saturation in spiking neural networks |
| CN108319146B (en) * | 2018-03-09 | 2020-08-11 | 西安西热控制技术有限公司 | Discrete particle swarm training-based method for radial basis function neural network |
| US11625609B2 (en) * | 2018-06-14 | 2023-04-11 | International Business Machines Corporation | Integration of external applications into deep neural networks |
| CN109978024B (en) * | 2019-03-11 | 2020-10-27 | 北京工业大学 | Effluent BOD prediction method based on interconnected modular neural network |
| CN111272969B (en) * | 2020-01-19 | 2022-02-22 | 西安热工研究院有限公司 | Method for predicting NOx generation concentration of 300MW pulverized coal boiler |
| CN114764093A (en) * | 2021-01-13 | 2022-07-19 | 新智数字科技有限公司 | Method and device for monitoring carbon monoxide content in flue gas of gas-fired boiler |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726481A (en) * | 2008-10-29 | 2010-06-09 | 北京华圣金程科技有限公司 | Method for identifying type of fuel on line |
| CN102830625A (en) * | 2012-09-10 | 2012-12-19 | 江苏科技大学 | Process control system and method based on neural network predictive control |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6736089B1 (en) * | 2003-06-05 | 2004-05-18 | Neuco, Inc. | Method and system for sootblowing optimization |
-
2014
- 2014-03-28 CN CN201410124690.1A patent/CN103870878B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726481A (en) * | 2008-10-29 | 2010-06-09 | 北京华圣金程科技有限公司 | Method for identifying type of fuel on line |
| CN102830625A (en) * | 2012-09-10 | 2012-12-19 | 江苏科技大学 | Process control system and method based on neural network predictive control |
Non-Patent Citations (1)
| Title |
|---|
| 基于神经网络的锅炉燃烧过程的建模分析;胡绍强;《中国优秀硕士学位论文全文数据库信息科技辑》;20130515;第2013年卷(第5期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103870878A (en) | 2014-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103870878B (en) | A kind of Combustion Characteristics in Utility Boiler neural network model | |
| CN103576655B (en) | A kind of power boiler burning subspace modeling and Multipurpose Optimal Method and system | |
| CN107726358B (en) | Boiler Combustion Optimization System based on CFD numerical simulations and intelligent modeling and method | |
| CN106931453B (en) | The forecasting system and method for circulating fluid bed domestic garbage burning emission of NOx of boiler | |
| CN102425807B (en) | Combustion feedforward and feedback composite optimization controlling method for pulverized coal fired boiler | |
| Hao et al. | Optimizing pulverized coal combustion performance based on ANN and GA | |
| CN102621945B (en) | Efficiency dynamic optimizing operation closed-loop optimization control method based on optimum operating conditions of thermal generator set | |
| CN105974793A (en) | Power plant boiler combustion intelligent control method | |
| CN103759290A (en) | Large coal-fired unit online monitoring and optimal control system and implementation method thereof | |
| CN104763999A (en) | Power plant pulverized coal boiler combustion performance online optimizing method and system | |
| CN103729691B (en) | Multi-target coal distribution method based on uniform design | |
| CN105020705A (en) | Method and system for optimizing and controlling combustion performance of circulating fluidized bed boiler in real time | |
| CN106527176A (en) | MFOA (modified fruit fly optimization algorithm)-SVM (support vector machine)-based boiler thermal efficiency and NOX modeling method | |
| CN103148506A (en) | Secondary air distribution method for pulverized coal boiler with swirling combustion of front-back hedging | |
| CN105240868B (en) | Boiler automatic combustion adjustment control method based on coal-air ratio coal correction | |
| CN105116730B (en) | Hydrogen-fuel engine electronic spark advance angle and optimizing system and its optimization method based on Particle Group Fuzzy Neural Network | |
| CN103886370B (en) | It is suitable for the power boiler burning performance neural network model of different coal pulverizer combination | |
| CN110084717A (en) | A kind of Utility Boiler moisture content of coal calculation method based on BP neural network | |
| CN103148507A (en) | Secondary air distribution method and system for pulverized coal boiler with swirling combustion of front-back hedging | |
| CN110017470A (en) | Boiler combustion multiobjective optimization control method based on differential evolution algorithm | |
| Ma et al. | Three-objective optimization of boiler combustion process based on multi-objective teaching–learning based optimization algorithm and ameliorated extreme learning machine | |
| Ye et al. | Multi-objective modeling of boiler combustion based on feature fusion and Bayesian optimization | |
| Liu et al. | Study on integrated simulation model of economic, energy and environment safety system under the low-carbon policy in Beijing | |
| CN106352339B (en) | A kind of gas-fired heater optimization of air-fuel ratio control system | |
| WO2020052413A1 (en) | Combustion control optimization method and apparatus for thermal generator sets and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |