CN104331736B - Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural - Google Patents

Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural Download PDF

Info

Publication number
CN104331736B
CN104331736B CN201410640074.1A CN201410640074A CN104331736B CN 104331736 B CN104331736 B CN 104331736B CN 201410640074 A CN201410640074 A CN 201410640074A CN 104331736 B CN104331736 B CN 104331736B
Authority
CN
China
Prior art keywords
rbf neural
dynamic
auxiliary variable
output
rbf
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
Application number
CN201410640074.1A
Other languages
Chinese (zh)
Other versions
CN104331736A (en
Inventor
张卫庆
李凡军
熊志化
高爱民
丁建良
柯炎
丁苏栋
殳建军
钱庆生
于国强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Corp of China SGCC
State Grid Jiangsu Electric Power Co Ltd
Jiangsu Fangtian Power Technology Co Ltd
Original Assignee
State Grid Corp of China SGCC
State Grid Jiangsu Electric Power Co Ltd
Jiangsu Fangtian Power Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by State Grid Corp of China SGCC, State Grid Jiangsu Electric Power Co Ltd, Jiangsu Fangtian Power Technology Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN201410640074.1A priority Critical patent/CN104331736B/en
Publication of CN104331736A publication Critical patent/CN104331736A/en
Application granted granted Critical
Publication of CN104331736B publication Critical patent/CN104331736B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

The invention discloses a kind of ultra-supercritical boiler NOx emission dynamic prediction methods based on RBF neural, belong to environment protection emission parameter measuring technical field.The present invention adopts the following technical scheme that:SS1 chooses static auxiliary variable and dynamic auxiliary variable;SS2 carries out RBF neural network structure fitting to the static auxiliary variable and the dynamic auxiliary variable, obtains the training emission of NOx of boiler dynamic prediction model based on RBF neural network structure;RBF neural parameter is adjusted, obtains the NOx emission dynamic prediction model of the ultra-supercritical boiler based on RBF networks.Under the conditions of identical training error etc. is set, the intrinsic nerve member number of dynamic model of the present invention is considerably less than static models, and model structure is simpler, and the training time is shorter, and generalization ability is stronger.

Description

Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural
Technical field
The present invention relates to the ultra-supercritical boiler NOx emission predictive method of current main-stream more particularly to based on RBF nerves The ultra-supercritical boiler NOx emission dynamic prediction method of network, belongs to environment protection emission parameter measuring technical field.
Background technology
With the development of ultra supercritical thermoelectricity technology and the increasingly complexity of power production process, it is ensured that process units stabilization, Efficiently, green operation, needs a pair significant process variable closely related with the safety in production, economical operation, environment protection emission of system Carry out monitoring and optimal control in real time.However, some important parameters are often in the ring of high temperature, high pressure, dusty, highly corrosive Border, it is difficult to directly measure.By taking coal-burning power plant's CEMS instrument as an example, when only having short-term by the data that sampling analysis device measures Inevitably there is measuring point no signal or measurement error beyond allowed band, need artificial regular in the characteristics of effect property, longtime running Inspection and maintenance (relevant industries standard regulation is calibrated daily with standard sample gas), probe timing purges, instrument periodic check etc., consumption A large amount of manpower and material resources.
Nitrogen oxides (NOx) pollution that fire coal generates is the primary goal that China's nitrogen oxides is administered.The denitration of coal dust is made In terms of with the same phase denitration including fugitive constituent and the out-phase denitration of coal tar two, mechanism is extremely complex.Meanwhile influence denitration efficiency Factor is also very much, including fire coal kind, temperature, residence time, reburning fuel ratio, oxygen concentration, fineness of pulverized coal, Secondary Air air distribution again Deng.Forecasting Methodology of the prior art chooses total fuel quantity, secondary air register aperture, CCOFA burnout degrees aperture, economizer exit oxygen (O2), coal characteristic parameter (Car, Har, Qar, Nar, Q, Var), coal pulverizer ventilation quantity, coal pulverizer coal-supplying amount are measured as auxiliary Variable.Qualified coal dust is worn into mill since fuel quantity instruction changes to feeder rotation speed change, raw coal, tube cell conveying, is arrived again Burner injection is one big inertia, Large Time Delay Process in the fever of stove chamber inner combustion, especially in the dynamic process of lifting load, The injection of burner moment has larger deviation in the coal dust in burner hearth and instantaneous coal-supplying amount.Instantaneous coal-supplying amount can not accurate characterization Into the instantaneous coal dust amount of burner hearth, choose instantaneous coal-supplying amount (or feeder aperture) and be unsatisfactory for the spirit that modeling auxiliary variable is chosen Quick property, accuracy requirement.
Neural network have powerful None-linear approximation performance, how by Radial Basis BP Neural Network (RBF) into The method of Mobile state modeling becomes the task of top priority among being applied to the prediction of discharged nitrous oxides.It is badly in need of a kind of according to model being estimated Meter or soft instrument solve the above problems.
Invention content
The present invention adopts the following technical scheme that:Ultra-supercritical boiler NOx emission dynamic prediction side based on RBF neural Method, which is characterized in that include the following steps:
SS1 chooses static auxiliary variable and dynamic auxiliary variable;
SS2 carries out RBF neural network structure fitting to the static auxiliary variable and the dynamic auxiliary variable, obtains base In the training emission of NOx of boiler dynamic prediction model of RBF neural network structure;The RBF neural includes input layer, implies Layer, output layer, the input layer include n sampled point, and the hidden layer includes N number of node, and the output layer includes 1 RBF Neural network exports, for arbitrary n sampled point { (xi,ti)|xi∈Rn,ti∈ R }, i=1,2 ..., n, structure are n-N-1's RBF neural exports
Wherein wj(j=1,2 ..., N) it is weights of j-th of hidden layer node to output node layer;φj(x) (j=1, 2 ..., N) gaussian kernel function for j-th hidden layer node, i.e.,
Wherein cjFor the data center of kernel function, σjIt is described implicit for all samples for the extension constant of the kernel function Node layer output matrix is
RBF neural output matrix form be
WhereinTo connect the weight matrix of hidden layer and output layer,Reality output matrix for RBF neural;
SS3 is adjusted RBF neural parameter, and the NOx emission for obtaining the ultra-supercritical boiler based on RBF networks is moved State prediction model, by RBF neural output valve Y and actual value T=(t1 t2 ... tn) between error sum of squares conduct The training objective error function of RBF neural, i.e.,
Make network output valve Y and actual value T=(t to find optimal output weights W1 t2 ... tn) between error put down Side and and output weights W Norm minimums, RBF neural parameter is optimized in two steps:Pass through the wide of matrix φ first The inverse optimal value for acquiring W of justice,
W=φ+T
Then using the error sum of squares of Y and T as object function, hidden node data center c is optimized by gradient descent algorithmj And extension constant σj, object function is to cjAnd σjGradient be respectively:
Data center cjAnd extension constant σjMore new formula be:
η is learning rate, k=1,2 ..., n.
Preferably, static auxiliary variable goes out including total fuel quantity, secondary air register aperture, CCOFA burnout degrees aperture, economizer Mouth oxygen amount, coal characteristic parameter, coal pulverizer ventilation quantity, dynamic auxiliary variable include total fuel quantity time delay elements one, total fuel quantity Time delay elements two, total fuel quantity time delay elements three.
The advantageous effect that the present invention is reached:(1) optimization algorithm that provides of the present invention, effectively merged static modelling and The advantages of dynamic modeling, chooses the time delay elements of important auxiliary variable (total fuel quantity) as dynamic auxiliary variable, passes through nerve Network is fitted, and has obtained the model for fully including ultra-supercritical boiler pulverized coal preparation system dynamic characteristic;(2) identical training is being set Under the conditions of error etc., the intrinsic nerve member number of dynamic model is considerably less than static models, and model structure is simpler, the training time Shorter, generalization ability is stronger.
Description of the drawings
Fig. 1 is the operational flow diagram of the present invention.
Fig. 2 is NOx content estimated value and assay value Comparative result curve graph under large sample.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention Technical solution, and be not intended to limit the protection scope of the present invention and limit the scope of the invention.
Subjects of the present invention become for Guohua Xuzhou Power Generation Co., Ltd. SG3099/27.46-M545 types ultra supercritical parameter Pressure operation spirally-wound tubes Once-through Boiler, rises to 1000MW, then the process of 500MW is down to by 1000MW in 24 hours by load 500MW Middle SCR Reactor inlets NOx content (butt standard state, 6%O2 states under).
Before model training is carried out, the invalid data collection that probe purge stages are locked first is weeded out, and data set is pressed Illuminated (12) is normalized in the range of [- 1,1], and renormalization is carried out according still further to formula (13) when model is exported and restored.
X'=(2x-xmax-xmin)/(xmax-xmin) (12)
X=[x'(xmax-xmin)+xmax+xmin]/2 (13)
Extensive mean square deviation εMSEIt calculates as follows
N in formulatestFor test sample number, t'iBe normalization after i-th of sample actual value, x'iIt is i-th after normalizing The predicted value of a sample.
Following scheme can be used in static state modeling method:Static scheme input includes:Total fuel quantity, secondary air register aperture (6 It is a), CCOFA burnout degrees aperture (4), oxygen content at economizer outlet, coal characteristic parameter (Car, Har, Qar, Nar, Q, Var), The instantaneous coal-supplying amount (6) of coal pulverizer ventilation quantity (6), coal pulverizer, the altogether input of 30 auxiliary variables.
The modeling method of the present invention uses:Total fuel quantity, secondary air register aperture (6), CCOFA burnout degrees aperture (4), Oxygen content at economizer outlet, coal characteristic parameter (Car, Har, Qar, Nar, Q, Var), coal pulverizer ventilation quantity (6), total fuel quantity Time delay elements one, total fuel quantity time delay elements two, total fuel quantity time delay elements three, the altogether input of 27 auxiliary variables.
Selecting identical 300 operating modes, identical 100 operating modes are as test sample, training as training sample Bias target is uniformly set as 0.3%.Shown in table one is dynamic model compared with the performance of two kinds of static models.
Table one
By table one it can be seen that the network structure that obtains of dynamic modelling method simplify, generalization ability it is strong.It is more preferable in order to obtain Extensive effect, increase training sample number to 1000, separately take 1000 as test sample, training bias target is still set as 0.3%.Shown in table two be under large sample dynamic model compared with the performance of static models.
Table two
Shown in Fig. 2 is that dynamic model estimation output valve (dotted line) and analysis meter actual measurement import NOx contain under large sample The Comparative result curve graph of (solid line) is measured, all experiments are completed under same PC, and allocation of computer is Core I5- 3230M, 4G RAM.Simulation software uses MATLAB R2014a, and operating system is WINDOWS 8.
NOx emission dynamic prediction can be implanted into industrial control process computer program by soft instrument mode.
The Major Difficulties of dynamic modeling are the selection of dynamic variable scheme and determining for dynamic characteristic time, can be with Time domain width and dynamic auxiliary variable distribution density are determined by gridding method optimizing.Pass through the emulation ratio with static state modeling method Compared with statistical data shows that dynamic modeling obtains preferable generalization ability with most succinct model structure, has higher engineering should With value.
The above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art For member, without departing from the technical principles of the invention, several improvement and deformation can also be made, these are improved and deformation Also it should be regarded as protection scope of the present invention.

Claims (1)

1. the ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural, which is characterized in that including walking as follows Suddenly:
SS1 chooses static auxiliary variable and dynamic auxiliary variable;The static state auxiliary variable includes total fuel quantity, secondary air register is opened Degree, CCOFA burnout degrees aperture, oxygen content at economizer outlet, coal characteristic parameter, coal pulverizer ventilation quantity, the dynamic auxiliary variable Including total fuel quantity time delay elements one, total fuel quantity time delay elements two, total fuel quantity time delay elements three;
SS2 carries out RBF neural network structure fitting to the static auxiliary variable and the dynamic auxiliary variable, is based on The training emission of NOx of boiler dynamic prediction model of RBF neural network structure;The RBF neural includes input layer, implies Layer, output layer, the input layer include n sampled point, and the hidden layer includes N number of node, and the output layer includes 1 RBF Neural network exports, for arbitrary n sampled point { (xi,ti)|xi∈Rn,ti∈ R }, i=1,2 ..., n, structure are n-N-1's RBF neural exports
Wherein wj(j=1,2 ..., N) it is weights of j-th of hidden layer node to output node layer;φj(x) (j=1,2 ..., N) the gaussian kernel function for j-th of hidden layer node, i.e.,
Wherein cjFor the data center of kernel function, σjFor the extension constant of the kernel function, for all samples, the hidden layer section Putting output matrix is
RBF neural output matrix form be
WhereinTo connect the weight matrix of hidden layer and output layer,Reality output matrix for RBF neural;
SS3 is adjusted RBF neural parameter, and the NOx emission dynamic for obtaining the ultra-supercritical boiler based on RBF networks is pre- Model is surveyed, by RBF neural output valve Y and actual value T=(t1 t2 ... tn) between error sum of squares as RBF god Training objective error function through network, i.e.,
Make network output valve Y and actual value T=(t to find optimal output weights W1 t2 ... tn) between error sum of squares And output weights W Norm minimums, RBF neural parameter is optimized in two steps:Pass through the generalized inverse of matrix φ first The optimal value of W is acquired,
W=φ+T
Then using the error sum of squares of Y and T as object function, hidden node data center c is optimized by gradient descent algorithmjAnd Extend constant σj, object function is to cjAnd σjGradient be respectively:
Data center cjAnd extension constant σjMore new formula be:
η is learning rate, k=1,2 ..., n.
CN201410640074.1A 2014-11-13 2014-11-13 Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural Active CN104331736B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410640074.1A CN104331736B (en) 2014-11-13 2014-11-13 Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410640074.1A CN104331736B (en) 2014-11-13 2014-11-13 Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural

Publications (2)

Publication Number Publication Date
CN104331736A CN104331736A (en) 2015-02-04
CN104331736B true CN104331736B (en) 2018-06-22

Family

ID=52406457

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410640074.1A Active CN104331736B (en) 2014-11-13 2014-11-13 Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural

Country Status (1)

Country Link
CN (1) CN104331736B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104763999A (en) * 2015-03-04 2015-07-08 内蒙古瑞特优化科技股份有限公司 Power plant pulverized coal boiler combustion performance online optimizing method and system
CN105629738B (en) * 2016-03-24 2018-06-29 内蒙古瑞特优化科技股份有限公司 SCR flue gas denitrification systems control method and equipment
CN106021916B (en) * 2016-05-18 2018-08-21 厦门大学 One kind being suitable for ultra-supercritical boiler NOxThe computational methods of discharge capacity analysis
CN108038561A (en) * 2017-09-21 2018-05-15 南京航空航天大学 A kind of Multipurpose Optimal Method of SCR denitration preformed catalyst
CN108956876B (en) * 2018-07-12 2020-12-29 浙江大学 Measurement delay correction method for smoke on-line continuous monitoring system
CN109187914B (en) * 2018-09-18 2020-12-25 哈尔滨锅炉厂有限责任公司 Coal quality characteristic-based method for predicting NOx generation amount of coal-fired power plant
CN110096785B (en) * 2019-04-25 2020-09-01 华北电力大学 Stack self-encoder modeling method applied to ultra-supercritical unit
CN112325329B (en) * 2020-10-13 2021-11-02 华中科技大学 High-temperature corrosion prevention boiler air door opening control method and system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103115356A (en) * 2013-02-07 2013-05-22 华北电力大学 Monitoring method and optimal control method for supercritical circulating fluidized bed (CFB) boiler combustion signals
CN103324862A (en) * 2013-07-11 2013-09-25 中国石油大学(华东) Coal-fired boiler optimization method based on improved neural network and genetic algorithm
CN103759290A (en) * 2014-01-16 2014-04-30 广东电网公司电力科学研究院 Large coal-fired unit online monitoring and optimal control system and implementation method thereof

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7899652B2 (en) * 2007-08-31 2011-03-01 Toyota Motor Engineering & Manufacturing North America, Inc. Linear programming support vector regression with wavelet kernel
CN102966945B (en) * 2012-11-02 2016-03-02 阳光凯迪新能源集团有限公司 Reduce the method for biomass recirculating fluidized bed boiler discharged nitrous oxides
CN103793746A (en) * 2014-02-19 2014-05-14 上海交通大学 Method for identifying parameters of coal-fired power plant boiler superheater model
CN103839110A (en) * 2014-02-24 2014-06-04 国家电网公司 Modeling method of prediction of production of nitrogen oxide in boiler

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103115356A (en) * 2013-02-07 2013-05-22 华北电力大学 Monitoring method and optimal control method for supercritical circulating fluidized bed (CFB) boiler combustion signals
CN103324862A (en) * 2013-07-11 2013-09-25 中国石油大学(华东) Coal-fired boiler optimization method based on improved neural network and genetic algorithm
CN103759290A (en) * 2014-01-16 2014-04-30 广东电网公司电力科学研究院 Large coal-fired unit online monitoring and optimal control system and implementation method thereof

Also Published As

Publication number Publication date
CN104331736A (en) 2015-02-04

Similar Documents

Publication Publication Date Title
CN104331736B (en) Ultra-supercritical boiler NOx emission dynamic prediction method based on RBF neural
Lv et al. A dynamic model for the bed temperature prediction of circulating fluidized bed boilers based on least squares support vector machine with real operational data
CN109388774A (en) A kind of thermal power plant NOx prediction model characteristics of variables extracting method based on method of comparison
CN111222284A (en) Overall soft measurement method for primary air volume at inlet of medium-speed coal mill unit
CN106202946A (en) Clinker free calcium levels Forecasting Methodology based on degree of depth belief network model
CN107103176A (en) Coal-burning boiler SCR catalyst life assessment method based on multisource information fusion technology
CN104715142B (en) A kind of station boiler NOxDischarge dynamic soft-measuring method
CN108954375B (en) Coal-saving control method for boiler
CN106649919A (en) Method and system for predicting carbon content of fly ash in coal-fired power plant boiler
CN110084717A (en) A kind of Utility Boiler moisture content of coal calculation method based on BP neural network
CN106327004A (en) Cement firing process optimizing method based on clinker quality index
Tan et al. NOx emission model for coal-fired boilers using principle component analysis and support vector regression
CN109785187B (en) Method for correcting power supply coal consumption detection data of generator set
CN110032747A (en) A kind of coal-burning boiler NOxDischarge On-line Measuring Method
CN110597070A (en) Method for identifying model parameters of thermal power generating unit system
Tunckaya et al. Comparative analysis and prediction study for effluent gas emissions in a coal-fired thermal power plant using artificial intelligence and statistical tools
Qin et al. A flame imaging-based online deep learning model for predicting NOₓ emissions from an oxy-biomass combustion process
CN105808945B (en) A kind of hybrid intelligent boiler efficiency burning optimization method
CN103823971A (en) Oxygen content detecting method for boiler fuel gas of coal-fired power station
Fu et al. Prediction model of desulfurization efficiency of coal-fired power plants based on long short-term memory neural network
CN110647560B (en) Mechanism and brain-like intelligence combined online soft measurement method for coal quality of fired coal of power plant
Agrawal et al. Application of K-NN regression for predicting coal mill related variables
Liangyu et al. ANN-based soft sensing of oxygen content in boiler air-flue gas system
CN112231970A (en) Boiler parameter modeling method based on time delay neural network
Xing et al. Dynamic soft sensing of organic pollutants in effluent from UMIC anaerobic reactor for industrial papermaking wastewater

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant