CN107694337A - Coal unit SCR denitrating flue gas control methods based on network response surface - Google Patents
Coal unit SCR denitrating flue gas control methods based on network response surface Download PDFInfo
- Publication number
- CN107694337A CN107694337A CN201711075034.7A CN201711075034A CN107694337A CN 107694337 A CN107694337 A CN 107694337A CN 201711075034 A CN201711075034 A CN 201711075034A CN 107694337 A CN107694337 A CN 107694337A
- Authority
- CN
- China
- Prior art keywords
- scr
- denitration system
- scr denitration
- flue gas
- network
- 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
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
- B01D—SEPARATION
- B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
- B01D53/34—Chemical or biological purification of waste gases
- B01D53/74—General processes for purification of waste gases; Apparatus or devices specially adapted therefor
- B01D53/86—Catalytic processes
- B01D53/90—Injecting reactants
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
- B01D—SEPARATION
- B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
- B01D53/34—Chemical or biological purification of waste gases
- B01D53/74—General processes for purification of waste gases; Apparatus or devices specially adapted therefor
- B01D53/86—Catalytic processes
- B01D53/8621—Removing nitrogen compounds
- B01D53/8625—Nitrogen oxides
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
- B01D—SEPARATION
- B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
- B01D53/34—Chemical or biological purification of waste gases
- B01D53/74—General processes for purification of waste gases; Apparatus or devices specially adapted therefor
- B01D53/86—Catalytic processes
- B01D53/8696—Controlling the catalytic process
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
- B01D—SEPARATION
- B01D2251/00—Reactants
- B01D2251/20—Reductants
- B01D2251/206—Ammonium compounds
- B01D2251/2062—Ammonia
Landscapes
- Engineering & Computer Science (AREA)
- Environmental & Geological Engineering (AREA)
- Chemical & Material Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Analytical Chemistry (AREA)
- General Chemical & Material Sciences (AREA)
- Oil, Petroleum & Natural Gas (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Treating Waste Gases (AREA)
Abstract
The present invention relates to a kind of coal unit SCR denitrating flue gas control methods based on network response surface, belong to gas denitrifying technology field.Comprise the following steps：Step S1, gathers the sample data on time change of SCR denitration system, and the input layer of dynamic neural network and the neuron of output layer are determined according to sample data；Step S2, Model Distinguish is carried out to SCR denitration system using dynamic neural network, establishes SCR forecast models；Step S3, the predicted value of the NOx concentration in SCR denitration system exit is calculated using SCR forecast models, and the ammonia spraying amount of SCR denitration system is controlled using the predicted value of the NOx concentration in SCR denitration system exit.Using method provided by the present invention outlet nitrous oxides concentration can be controlled to keep constant substantially, it can accurately meet to require in real time in ammonia spraying amount formulation, the problem of wasting reducing agent, the escaping of ammonia increase is overcome, substantially increases the accuracy of ammonia spraying amount PREDICTIVE CONTROL.
Description
Technical field
The invention belongs to gas denitrifying technology field, is related to a kind of coal unit SCR based on network response surface
Denitrating flue gas control method.
Background technology
SCR (Selective Catalytic Reduction) --- selective catalytic reduction is current skill in the world
Art is most ripe, most widely used gas denitrifying technology.SCR is in the presence of catalyst, utilizes reducing agent NH3Etc. there is a choosing
Reacted with the NOx in flue gas to selecting property and generate the N of nontoxic pollution-free2And H2O.Control effect of the SCR technology to boiler smoke NOx
Fruit is very notable, and technology is more ripe, has turned at present in the world using a kind of most, most fruitful gas denitrifying technologies.
In denitrating system, ammonia flow is to be multiplied by NH by NOx flow signals3/ NOx mol ratios obtain, wherein,
NH3/ NOx mol ratios are fixed, and NOx flows are smoke inlet NOx concentration and the product of flue gas flow.Exhanst gas outlet simultaneously
NOx concentration is to NH3Demand is corrected, and finally draws required ammonia flow value.Denitration control system is according to calculating
The ammonia flow value gone out, by controlling ammonia to adjust the aperture of door, it is possible to achieve ammonia flow automatically controls.
Ammonia spraying amount is the important controlled quentity controlled variable of SCR denitration system, and power plant generally uses PID control ammonia spraying amount at present, works as unit
Load in stable state, preferable control effect can be obtained, but under the conditions of variable working condition, system shows non-linear, big
Hysteresis quality, it is difficult to ensure most preferably to spray ammonia ratio.If ammonia spraying amount is very few, NOx emission standard is difficult to ensure that；If ammonia spraying amount mistake
It is more, then the waste of ammonia is not only caused, and new pollution can be caused again.Therefore, often it is difficult to using the pid control mode of routine
Obtain preferable control effect.
The content of the invention
The present invention provides a kind of coal unit SCR denitrating flue gas control methods based on network response surface, with solution
Certainly because causing to be difficult to ensure that NOx emission standard can not accurately controlling the ammonia spraying amount of the out of stock systems of SCR under the conditions of variable working condition, wasting
The problem of reducing agent, increase the escaping of ammonia.
The technical solution adopted by the present invention is to comprise the following steps：
Step S1, the sample data on time change of SCR denitration system is gathered, and determined according to the sample data
The input layer of dynamic neural network and the neuron of output layer；
Step S2, Model Distinguish is carried out to the SCR denitration system using the dynamic neural network, establishes SCR predictions
Model；
Step S3, the predicted value of the NOx concentration in the SCR denitration system exit is calculated using the SCR forecast models,
And the ammonia spraying amount of the SCR denitration system is controlled using the predicted value of the NOx concentration in the SCR denitration system exit
System.
At the NOx concentration of sample data of the present invention including the SCR denitration system porch, SCR reaction units
Temperature, unit load, the ammonia spraying amount of the SCR denitration system and the NOx concentration in the SCR denitration system exit.
Before the step S2, methods described also includes the present invention：Step S4, high frequency filter is carried out to the sample data
Ripple processing；Step S5, the sample data after progress High frequency filter processing is normalized.
The neuron of the input layer of dynamic neural network of the present invention is dense for the NOx of the SCR denitration system porch
The ammonia spraying amount of temperature, unit load and the SCR denitration system at degree, SCR reaction units；The dynamic neural network it is defeated
The neuron for going out layer is the NOx concentration in the SCR denitration system exit.
The present invention gather the frequency of the sample data for it is per minute once.
Dynamic neural network of the present invention is made up of BP neural network.
After the Model Distinguish in the step S2, the delay for determining the input layer of the BP neural network is the present invention
In 10 sampling periods, the delay of output layer is 2 sampling periods, and the number of hidden layer is 60, the activation primitive of the hidden layer
For non-negative logarithm S function, i.e. logsig (x).
The hidden layer of BP neural network of the present invention uses Sigmoid functions, i.e.,Profit
The NOx concentration in SCR denitration system exit optimizes described in the Sigmoid function pairs, by continuous iteration, using most
Fast gradient method tries to achieve optimal ammonia spraying amount.
The present invention establishes P BP neural network when controlling time domain to be walked for P according to time sequencing, wherein, s-th of BP god
Through network B PsIt is expressed as：
Footmark s represents s-th of BP network, x in formulaiFor the input of i-th of hidden node, corresponding output is zi,Represent defeated
Ingress j to hidden node i link weight coefficients,Hidden node i Input Offset Value is represented,For the input offset of input node
Value.
The present invention has advantages below：
1st, principal element is extracted from many factors for influenceing SCR denitration efficiency, sample data is obtained according to principal element,
The sample data has the characteristics of ergodic, compactness and compatibility, beneficial to simplified system model；
2nd, the present invention is dense according to the nitrous oxides concentration of Neural Network model predictive and the nitrogen oxides of SCR system porch
Spend to control the ammonia spraying amount in SCR denitration system, specifically, the sample data on time change of collection SCR denitration system
To carry out SCR denitration system Model Distinguish, and model pre-estimating is carried out to the system using non linear autoregressive model, rather than it is single
Pure carries out PID control (technology that current domestic power plant uses) to export nitrous oxides concentration as deviation, can control substantially
Make mouthful nitrous oxides concentration and keep constant, tail gas is reached expected limit row's standard, can be accurate in real time in ammonia spraying amount formulation
True satisfaction requirement, overcomes the problem of wasting reducing agent, the escaping of ammonia increase, substantially increases the accurate of ammonia spraying amount PREDICTIVE CONTROL
Property, contrasted using the sample and simulation result of collection, accuracy can reach 0.99；
3rd, for postponing the out of stock systems of SCR larger, that disturbance is more, excellent effect, the robust of present invention control spray according to quantity
Property it is good, possess adaptive, self study, the ability of Self-tuning System, and regulating time is shorter, dynamic error is smaller；
4th, it is relatively low to model needs, be easy to preferable in line computation, control effect using the Prediction and Control Technology of the present invention.
Brief description of the drawings
Fig. 1 is the coal unit SCR denitrating flue gas control provided in an embodiment of the present invention based on network response surface
The flow chart of method；
Fig. 2 is input sample data original in the embodiment of the present invention and time T curve map；
Fig. 3 is output sample data original in the embodiment of the present invention and time T curve map；
Fig. 4 is the filtered input sample data of medium-high frequency of the embodiment of the present invention and time T curve map；
Fig. 5 is the filtered output sample data of medium-high frequency of the embodiment of the present invention and time T curve map；
Fig. 6 is that the NOx in the SCR denitration system exit being calculated in the embodiment of the present invention according to SCR forecast models is dense
The predicted value of degree and the comparison diagram of actual value；
Fig. 7 A to 7D are respectively training data in the embodiment of the present invention, verification data, test data and triplicity
The linearity of regression curve map of predicted value and actual value；
Fig. 8 is Model Predictive Control frame diagram in the embodiment of the present invention；
Fig. 9 is the PREDICTIVE CONTROL block diagram of neutral net in the embodiment of the present invention；
Figure 10 is control system block diagram in the embodiment of the present invention；
Figure 11 is preceding 50 step PREDICTIVE CONTROL analogous diagram in the embodiment of the present invention.
Embodiment
Comprise the following steps：
Step S1, the sample data on time change of SCR denitration system is gathered, and dynamic is determined according to sample data
The input layer of neutral net and the neuron of output layer；
In the present embodiment, in order to provide the input and the output number that comprehensively, correctly reflect system performance to prototype network
According to right, the data for gathering and putting into network training should meet following three points characteristic：Ergodic, compactness and compatibility.Learn
Sample includes object all state space situations that may be present as far as possible；And the learning sample in certain spatial dimension
Density is suitable, plant characteristic could so be collected.The sample size that training network needs depends on answering for research object
Miscellaneous degree and noise to the influence degree of experimental subjects, i.e., system complexity and noise level determine needed for sample size it is big
It is small.In the case where there is big sample size, it is contemplated that the state that is likely to occur of system just merge it is better, at this moment accordingly
Deviation may be reduced.When sample has arrived certain amount, the effect that it improves system model is more and more weaker.Sample it is compatible
Property refers to that in the higher space of overlapping degree very much like input often corresponds to completely different output.
Having obtained some by site monitoring system mainly influences the factor of the out of stock efficiency of SCR：Ammonia flow, reactor enter
Mouth amount of nitrogen oxides, generator active power, temperature of reactor, reactor outlet amount of nitrogen oxides, therefore, SCR is taken off
Temperature, unit load at the NOx concentration of nitre system entrance, SCR reaction units, the ammonia spraying amount and SCR of SCR denitration system
The NOx concentration in denitrating system exit is as sample data.
In the present embodiment, by the temperature at the NOx concentration of SCR denitration system porch, SCR reaction units, unit load
With 4 neurons of the ammonia spraying amount of SCR denitration system as the input layer of dynamic neural network；And SCR denitration system is exported
Neuron of the NOx concentration at place as the output layer of dynamic neural network.
Because to time frequency range, without specific requirement, but the object is process control, process variable change is slow, therefore in reality
Power plant experiment in the frequency of collecting sample data can be set to it is per minute once, i.e., it is per minute collection data.
As a kind of optional embodiment of the present embodiment, dynamic god is determined according to the sample data of collection in step s 2
Before the neuron that input layer through network includes, this method also includes：
Step S4, High frequency filter processing is carried out to sample data；
The data that site monitoring system is measured are divided into A groups and B groups, it is contemplated that denitration unit symmetric property, therefore herein
The situation of A sides is only discussed, is shown in Table 1.
The part gathered data of table 1
Draw original input sample data (A reactor inlets NOx as shown in Figure 2 respectively according to the gathered data of table 1
Concentration, generator active power, the out of stock temperature of reactor in A sides and A sides ammonia flow) and original output as shown in Figure 3
Sample data (A reactor outlets NOx concentration) and time T curve map.As shown in Figure 2, the out of stock temperature of reactor in A sides not with
Change too greatly occurs for time T changes.As shown in Figure 3：There is saturated phenomenon in A reactor outlet NOx concentrations, i.e., when ammonia spraying amount reaches
During to certain upper limit, the NOx concentration no longer step-down of SCR denitration system output, i.e., having some in ammonia spraying amount now is escaped with ammonia
What the form of ease was gone out.
Figure it is seen that the sample data of collection in worksite has some high-frequency signals, first these sample datas should be entered
Row High frequency filter.Specifically, the gap of more two neighboring sample data, if gap is more than certain value, (occurrence is according to needed for
Depending on the high-frequency signal scope filtered), then the data of later moment in time are filtered, and the moment value is equal to previous moment value.Fig. 4 is
Input sample data (A reactor inlets NOx concentration, generator active power, the out of stock temperature of reactor in A sides after High frequency filter
With A sides ammonia flow) with time T curve map, Fig. 5 it is that (A reactor outlets NOx is dense for output sample data after High frequency filter
Degree) with time T curve map.
Some high-frequency signals existing for scene can be avoided to improve the standard for modeling input data to signal collected interference
True property, and then finally improve the accuracy of exit ammonia spraying amount prediction.
Step S5, the sample data after progress High frequency filter processing is normalized.
Specifically, temperature, unit load and the SCR at the NOx concentration of SCR denitration system porch, SCR reaction units take off
The dimension and fluctuation range of this four variables of the ammonia spraying amount of nitre system are different, it is necessary to be normalized, and remove dimension, will
This four variables are unified onto an order of magnitude, can represent under same coordinate, therefore, can effectively be carried in training network
The fitting precision of high training speed and network.
Step S2, Model Distinguish is carried out to SCR denitration system using dynamic neural network, establishes SCR forecast models；It is optional
, dynamic neural network can be made up of BP neural network.The validity function of BP learning algorithms is mean square error, that is, not
Disconnectedly network is exported and made comparisons with reality output, equivalent to there is target, the just inspection that then often makes a move deviates target trajectory
Degree, then based on current location in the case of, adjust the route of next step.It is actual for each input sample, network
Output is just compared with desired output.
After Model Distinguish, the delay for determining the input layer of BP neural network is 10 sampling periods, and output layer prolongs
When be 2 sampling periods, the number of hidden layer is 60, and the activation primitive of hidden layer is non-negative logarithm S function, i.e. logsig (x).
Unlimited using neutral net approaches non-linear behaviour, and mould is carried out to SCR denitration system using dynamic neural network
Type recognizes, so as to accurate founding mathematical models, so as to reduce deviation existing for control system practical operation situation.
Step S3, the predicted value of the NOx concentration in SCR denitration system exit is calculated using SCR forecast models, and is utilized
The predicted value of the NOx concentration in SCR denitration system exit is controlled to the ammonia spraying amount of SCR denitration system.
Fig. 6 is the predicted value and reality of the NOx concentration in the SCR denitration system exit being calculated according to SCR forecast models
The comparison diagram of actual value.It will be appreciated from fig. 6 that the predicted value and actual value that are obtained using the SCR forecast models described in the present embodiment are basic
Overlapping, predicted value is relatively accurate, mean square deviation 0.7553, can be a little bigger although deviateing in data jump larger part error, but still
So it is more or less the same.
It should be noted that the abscissa unit in Fig. 2 to Fig. 6 is s, and ordinate dimensionless.
In order to further analyze the performance of the SCR forecast models of the present embodiment, training data, check number are depicted
According to, test data and the predicted value of triplicity and the linearity of regression curve map of actual value, respectively such as Fig. 7 A, 7B, 7C, 7D institute
Show.As shown in Figure 7, either obtained for training data, test data, verification data using the SCR forecast models of the present embodiment
For the prediction data and the similarity of real data arrived all more than 0.99, overall similarity reaches 0.996, therefore the spray ammonia predicted
The accuracy rate of amount is higher.
In PREDICTIVE CONTROL effect, future is exported and carries out multi-step prediction, i.e., when it is P steps to control time domain, according to the time
Order establishes P BP neural network, wherein, s-th of BP neural network BPsIt is expressed as：
Footmark s represents s-th of BP network, x in formulaiFor the input of i-th of hidden node, corresponding output is zi,Represent defeated
Ingress j to hidden node i link weight coefficients,Hidden node i Input Offset Value is represented,For the input offset of input node
Value.
The operation principle of this P neutral net is identical, is all the network established using nonlinear auto-companding principle.Difference
It is in and staggers in succession in time in their input quantity, can so makes the following output prediction at different moments of network output reflection
Value.And these networks can be carried out parallel in learning process and real-time estimate, therefore this method is feasible, and largely effective.
Wherein, the hidden layer of BP neural network uses Sigmoid functions (action function of neuron), is declined using gradient
Method (used step-length is 0.5) obtains Sigmoid functions, i.e.,Now, the model of foundation is more
It is accurate to add, and simulated effect is more preferable；Optimized, passed through using the NOx concentration in Sigmoid function pair SCR denitration systems exit
Continuous iteration, optimal ammonia spraying amount is tried to achieve using steepest gradient method, so as to effectively solve SCR denitration system it is big delay, it is non-thread
Property the problem of, for SCR control optimization new way is provided.
The mapping relations of input and output can be obtained according to SCR forecast models.By constantly solving cost function forward
Minimum value, i.e., at each moment, a nonlinear optimal problem is solved, next control action amount is obtained, by that analogy, obtains
Entirely desired control domain.The neural network model that the present embodiment is established, there is two effects, when as forecast model, second,
Its inverse mapping can as performance function J (k) optimization according to.
Wherein, it can represent as follows in moment k optimality criterion J (k)：
In formulaIt is the prediction output valve of each BP forecast models, they are that input will be u (k in future
+ h-1) (h=1 ..., P) when output valve, yr(k+h) (h=1 ..., P) it is output desired value.
Target capabilities function pair input value u (k+h-1) (h=1 ..., P) is sought into local derviation：
In formula (3), work as s<During hIt is unrelated with u (k+h-1), then, it can be write as：
In formula (4) on the right side of equationIt can be obtained according to formula (2)
And in formula (4) on the right side of equationIt can be tried to achieve by (1) formula
Finally have
So, can be with one group of controlled quentity controlled variable u of initial setting upm(k), calculated using model (1)Then performance is brought into
Function (2) is calculated in J (k)On this basis, controlled quentity controlled variable is improved with gradient method
Wherein α is step-length, and Grad can be calculated according to formula (7).This iterative process is repeated, until
To the J (k) of minimum, u (k) at this moment can be applied in system as optimum control amount to be implemented.
In addition, in order to verify the present invention, inventor has done following work：
1. the PREDICTIVE CONTROL that designs a model framework, as shown in figure 8, v represents the measurable disturbance in real system, this is not wish
Hope the variable for ringing output；R is a setting value, represents the desired value of output desired value, i.e. system middle outlet nitrogen oxides；U is
Operable variable, the amount of ammonia is as inputted, its size can be adjusted by controller, so as to act on destination object, is made
Output reaches desired value；D is not measurable disturbance, directly acts on target, has a great influence；Y is the output of measurement, i.e., is surveyed in system
The outlet nitrous oxides concentration of amount, it can be estimated that whether is the accuracy of real output value；Z is the noise for influenceing measurement accuracy；Real output value, i.e., the actual value of the NOx concentration at the described out of stock system outlets of SCR；MPC(ModelPredictive
Controller) it is model predictive controller.
2. the network response surface block diagram based on above-mentioned Frame Design, as shown in figure 9, including reference model
(Reference Model), cost function minimization (Cost Function Minimization), selector (Switch),
Neural plant model (Neural Plant Model), controlled device (Plant).
3. on the basis of based on network response surface principle, control system block diagram is built, as shown in Figure 10.System
System block diagram is mainly made up of three parts：Signal input (i.e. input time sequence), control optimization part, SCR denitration system part.
Wherein input time sequence is using the data collected as signal；Optimization part use gradient descent method (use step-length for
0.5) S function, has been write, S function is substantially carried out the optimization of exit NOx concentration, by continuous iteration, utilizes
Steepest gradient method tries to achieve optimal ammonia spraying amount；SCR denitration system is replaced by obtained neural network model.Putting up system frame
After figure, simulated environment is set.Here it is 50s to set simulation time, and simulation step length is fixed step size 1.0s, read access time sequence after every 1s
Data in row, the continuous propulsion of passage time sequence, the output of NOx concentration is tried to achieve, solver is ode3 (Bogacki-
Shampine)。
4. in order to obtain comparatively ideal data, the iterations set in S function is longer.Analyze 50 data, Figure 11
For preceding 50 step PREDICTIVE CONTROL analogous diagram.As can be seen from Figure 11, although having, some NOx exit concentrations are unstable, larger, big portion of jumping
NOx exit concentrations can be tracked near setting value between timesharing；In preceding 20 step, control effect is preferable；It is longer in prediction time domain
Afterwards, control effect is undesirable；So with reference to PID control, when entrance NOx changes more violent, a part of NOx is first removed with feedforward
Material, such control effect can be better.
First to SCR denitration system, this non-linear object carries out Model Distinguish to the present invention, using non linear autoregressive model
Model pre-estimating is carried out to the system, and then ammonia spraying amount is controlled using PREDICTIVE CONTROL thought, tail gas is reached limit row
Standard, it can also reduce with ammonia amount, reduction the escaping of ammonia, lifting economic benefit.Device is wherein controlled using steepest gradient method
Optimization, and controlled quentity controlled variable is constrained by performance function, reach anticipated output.Simulation result finally is surveyed into data with scene to carry out
Contrast, the results showed that network response surface scheme can accurately predict the ammonia spraying amount needed for following Finite time.
Forecast model is established to SCR denitration process and substitutes into the feasibility test that test data carries out the model.Prediction result is divided
Analysis, obtains response of the system in different time sections.
Claims (9)
1. a kind of coal unit SCR denitrating flue gas control methods based on network response surface, it is characterised in that including under
Row step：
Step S1, the sample data on time change of SCR denitration system is gathered, and dynamic is determined according to the sample data
The input layer of neutral net and the neuron of output layer；
Step S2, Model Distinguish is carried out to the SCR denitration system using the dynamic neural network, establishes SCR forecast models；
Step S3, the predicted value of the NOx concentration in the SCR denitration system exit, and profit are calculated using the SCR forecast models
The ammonia spraying amount of the SCR denitration system is controlled with the predicted value of the NOx concentration in the SCR denitration system exit.
2. the coal unit SCR denitrating flue gas control methods according to claim 1 based on network response surface, its
It is characterised by, temperature at the NOx concentration of the sample data including the SCR denitration system porch, SCR reaction units,
The NOx concentration in unit load, the ammonia spraying amount of the SCR denitration system and the SCR denitration system exit.
3. the coal unit SCR denitrating flue gas controlling parties according to claim 1 or 2 based on network response surface
Method, it is characterised in that before the step S2, in addition to：
Step S4, High frequency filter processing is carried out to the sample data；
Step S5, the sample data after progress High frequency filter processing is normalized.
4. the coal unit SCR denitrating flue gas controls based on network response surface according to any one of claims 1 to 3
Method processed, it is characterised in that：The neuron of the input layer of the dynamic neural network is the SCR denitration system porch
The ammonia spraying amount of temperature, unit load and the SCR denitration system at NOx concentration, SCR reaction units；The dynamic neural net
The neuron of the output layer of network is the NOx concentration in the SCR denitration system exit.
5. the coal unit SCR denitrating flue gas controls based on network response surface according to any one of Claims 1-4
Method processed, it is characterised in that：Gather the frequency of the sample data for it is per minute once.
6. the coal unit SCR denitrating flue gas controls based on network response surface according to any one of claim 1 to 5
Method processed, it is characterised in that：The dynamic neural network is made up of BP neural network.
7. the coal unit SCR denitrating flue gas control methods according to claim 6 based on network response surface, its
It is characterised by：After the Model Distinguish in the step S2, the delay for determining the input layer of the BP neural network is 10
In the sampling period, the delay of output layer is 2 sampling periods, and the number of hidden layer is 60, and the activation primitive of the hidden layer is non-
Negative logarithm S function, i.e. logsig (x).
8. the coal unit SCR denitrating flue gas control methods according to claim 7 based on network response surface, its
It is characterised by：The hidden layer of the BP neural network uses Sigmoid functions, i.e.,β>0；Using described
The NOx concentration in SCR denitration system exit optimizes described in Sigmoid function pairs, by continuous iteration, utilizes steepest gradient
Method tries to achieve optimal ammonia spraying amount.
9. the coal unit SCR denitrating flue gas control methods according to claim 8 based on network response surface, its
It is characterised by, when controlling time domain to be walked for P, P BP neural network is established according to time sequencing, wherein, s-th of BP neural network
BPsIt is expressed as：
Footmark s represents s-th of BP network, x in formulaiFor the input of i-th of hidden node, corresponding output is zi,Represent input section
Point j to hidden node i link weight coefficients,Hidden node i Input Offset Value is represented,For the Input Offset Value of input node.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711075034.7A CN107694337A (en) | 2017-11-03 | 2017-11-03 | Coal unit SCR denitrating flue gas control methods based on network response surface |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711075034.7A CN107694337A (en) | 2017-11-03 | 2017-11-03 | Coal unit SCR denitrating flue gas control methods based on network response surface |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107694337A true CN107694337A (en) | 2018-02-16 |
Family
ID=61177865
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711075034.7A Pending CN107694337A (en) | 2017-11-03 | 2017-11-03 | Coal unit SCR denitrating flue gas control methods based on network response surface |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107694337A (en) |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109046021A (en) * | 2018-07-19 | 2018-12-21 | 南京奥麦科仪器自动化有限公司 | A kind of SCR system of strong adaptive ability accurately sprays ammonia control method |
| CN109107360A (en) * | 2018-11-01 | 2019-01-01 | 国家能源投资集团有限责任公司 | Spray ammonia control method and device based on denitrating system |
| CN109190848A (en) * | 2018-10-17 | 2019-01-11 | 大唐环境产业集团股份有限公司 | A kind of SCR system NO based on Time-delay PredictionxConcentration of emission prediction technique |
| CN109304086A (en) * | 2018-10-31 | 2019-02-05 | 华中科技大学 | A kind of station boiler SCR denitration fining spray ammonia control method |
| CN109343367A (en) * | 2018-10-26 | 2019-02-15 | 齐鲁工业大学 | A method of based on network response surface flue gas desulfurization |
| CN109343349A (en) * | 2018-11-01 | 2019-02-15 | 大唐环境产业集团股份有限公司 | A kind of SCR denitrating flue gas Optimal Control System and method based on ammonia spraying amount compensator |
| CN109766596A (en) * | 2018-12-25 | 2019-05-17 | 国网新疆电力有限公司电力科学研究院 | A kind of expert system construction method of denitration economical operation |
| CN109876658A (en) * | 2019-04-08 | 2019-06-14 | 国电环境保护研究院有限公司 | Subregion vortex volume inhales spray ammonia system and spray ammonia adjusting method |
| CN109933884A (en) * | 2019-03-08 | 2019-06-25 | 华北电力大学（保定） | A method of towards coal unit SCR denitration system Neural network inverse control |
| CN110026068A (en) * | 2019-04-08 | 2019-07-19 | 东南大学 | A kind of large-scale coal fired power plant CO based on Neural network inverse control2Trapping system and feed forward control method |
| CN110299188A (en) * | 2019-05-09 | 2019-10-01 | 上海电力学院 | SCR flue gas denitrification system GRNN modeling method based on GA variables choice |
| CN110908351A (en) * | 2019-11-25 | 2020-03-24 | 东南大学 | Support vector machine-fused SCR denitration system disturbance suppression prediction control method |
| CN110935312A (en) * | 2019-12-16 | 2020-03-31 | 广州珠江电力有限公司 | Dynamic monitoring device and dynamic monitoring method for SCR flue gas denitration system |
| CN111413938A (en) * | 2020-04-16 | 2020-07-14 | 南京英璞瑞自动化科技有限公司 | SCR denitration system disturbance suppression prediction control method based on converted ammonia injection amount |
| CN111589301A (en) * | 2020-05-29 | 2020-08-28 | 广东电科院能源技术有限责任公司 | Method, device, equipment and storage medium for predicting SCR denitration performance of coal-fired power plant |
| CN111897373A (en) * | 2020-08-05 | 2020-11-06 | 海南创实科技有限公司 | Model prediction-based ammonia injection flow adjusting method for SCR denitration device |
| CN112180731A (en) * | 2020-10-13 | 2021-01-05 | 天津大学 | Energy equipment operation control method and system |
| CN112461995A (en) * | 2020-11-03 | 2021-03-09 | 西安热工研究院有限公司 | Method for predicting ammonia escape of thermal power plant |
| CN112488145A (en) * | 2019-11-26 | 2021-03-12 | 大唐环境产业集团股份有限公司 | NO based on intelligent methodxOnline prediction method and system |
| CN112596474A (en) * | 2020-11-23 | 2021-04-02 | 山东微立方信息技术股份有限公司 | Uniformity optimization method for SCR denitration system |
| CN112619394A (en) * | 2020-11-24 | 2021-04-09 | 呼和浩特科林热电有限责任公司 | Denitration ammonia injection self-adaptive control method and device and denitration system |
| CN112627953A (en) * | 2020-12-22 | 2021-04-09 | 上海海事大学 | Ship SCR system ammonia injection amount control method based on ARMAX and MMPC |
| CN112651166A (en) * | 2020-11-24 | 2021-04-13 | 呼和浩特科林热电有限责任公司 | Denitration system inlet nitrogen oxide concentration prediction method and device and denitration system |
| CN112933913A (en) * | 2021-02-08 | 2021-06-11 | 国家能源集团国源电力有限公司 | Ammonia injection control method and device and coal combustion system |
| CN113144843A (en) * | 2021-03-16 | 2021-07-23 | 国家能源集团国源电力有限公司 | Ammonia spraying control method and device |
| CN113398758A (en) * | 2021-07-20 | 2021-09-17 | 山信软件股份有限公司 | Ammonia water spraying flow control system and control method thereof |
| CN113488111A (en) * | 2021-07-05 | 2021-10-08 | 东南大学 | Ammonia injection amount optimization modeling method for SCR denitration system |
| CN113996158A (en) * | 2021-08-16 | 2022-02-01 | 国能(山东)能源环境有限公司 | Control method and device for external desulfurization, electronic equipment and storage medium |
| CN114326375A (en) * | 2021-12-22 | 2022-04-12 | 江苏精瓷智能传感技术研究院有限公司 | Nitrogen oxygen sensor pump unit self-adaptive control system |
| CN112180731B (en) * | 2020-10-13 | 2024-05-31 | 天津大学 | Energy equipment operation control method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104504238A (en) * | 2014-11-21 | 2015-04-08 | 国家电网公司 | Method for predicting nitrogen oxide emission concentration of coal-fired unit based on neural network |
| CN104826492A (en) * | 2015-04-23 | 2015-08-12 | 华北电力大学（保定） | Improvement method for selective catalytic reduction flue gas denitrification and ammonia injection control system |
| CN105629738A (en) * | 2016-03-24 | 2016-06-01 | 内蒙古瑞特优化科技股份有限公司 | SCR (Selective Catalytic Reduction) flue gas denitration system control method and apparatus |
| US20160310896A1 (en) * | 2015-04-23 | 2016-10-27 | Rockwell Automation Technologies, Inc. | Predective emissions monitor systems and methods |
-
2017
- 2017-11-03 CN CN201711075034.7A patent/CN107694337A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104504238A (en) * | 2014-11-21 | 2015-04-08 | 国家电网公司 | Method for predicting nitrogen oxide emission concentration of coal-fired unit based on neural network |
| CN104826492A (en) * | 2015-04-23 | 2015-08-12 | 华北电力大学（保定） | Improvement method for selective catalytic reduction flue gas denitrification and ammonia injection control system |
| US20160310896A1 (en) * | 2015-04-23 | 2016-10-27 | Rockwell Automation Technologies, Inc. | Predective emissions monitor systems and methods |
| CN105629738A (en) * | 2016-03-24 | 2016-06-01 | 内蒙古瑞特优化科技股份有限公司 | SCR (Selective Catalytic Reduction) flue gas denitration system control method and apparatus |
Non-Patent Citations (1)
| Title |
|---|
| 孟范伟 等: "神经网络预测控制在SCR烟气脱硝系统中应用", 《东北大学学报（自然科学版）》 * |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109046021A (en) * | 2018-07-19 | 2018-12-21 | 南京奥麦科仪器自动化有限公司 | A kind of SCR system of strong adaptive ability accurately sprays ammonia control method |
| CN109046021B (en) * | 2018-07-19 | 2021-03-02 | 南京奥麦科仪器自动化有限公司 | SCR system accurate ammonia injection control method with strong self-adaptive capacity |
| CN109190848B (en) * | 2018-10-17 | 2024-06-07 | 大唐环境产业集团股份有限公司 | SCR system NO based on time delay predictionxEmission concentration prediction method |
| CN109190848A (en) * | 2018-10-17 | 2019-01-11 | 大唐环境产业集团股份有限公司 | A kind of SCR system NO based on Time-delay PredictionxConcentration of emission prediction technique |
| CN109343367A (en) * | 2018-10-26 | 2019-02-15 | 齐鲁工业大学 | A method of based on network response surface flue gas desulfurization |
| CN109304086A (en) * | 2018-10-31 | 2019-02-05 | 华中科技大学 | A kind of station boiler SCR denitration fining spray ammonia control method |
| CN109304086B (en) * | 2018-10-31 | 2020-11-24 | 华中科技大学 | Power station boiler SCR denitration refined ammonia injection control method |
| CN109107360B (en) * | 2018-11-01 | 2021-09-21 | 国家能源投资集团有限责任公司 | Ammonia injection control method and device based on denitration system |
| CN109343349B (en) * | 2018-11-01 | 2023-08-22 | 大唐环境产业集团股份有限公司 | SCR flue gas denitration optimal control system and method based on ammonia injection amount compensator |
| CN109343349A (en) * | 2018-11-01 | 2019-02-15 | 大唐环境产业集团股份有限公司 | A kind of SCR denitrating flue gas Optimal Control System and method based on ammonia spraying amount compensator |
| CN109107360A (en) * | 2018-11-01 | 2019-01-01 | 国家能源投资集团有限责任公司 | Spray ammonia control method and device based on denitrating system |
| CN109766596A (en) * | 2018-12-25 | 2019-05-17 | 国网新疆电力有限公司电力科学研究院 | A kind of expert system construction method of denitration economical operation |
| CN109933884A (en) * | 2019-03-08 | 2019-06-25 | 华北电力大学（保定） | A method of towards coal unit SCR denitration system Neural network inverse control |
| CN109933884B (en) * | 2019-03-08 | 2023-08-29 | 华北电力大学（保定） | Neural network inverse control method for SCR denitration system of coal-fired unit |
| CN109876658B (en) * | 2019-04-08 | 2023-07-07 | 国电环境保护研究院有限公司 | Partition vortex entrainment ammonia spraying system and ammonia spraying adjusting method |
| CN110026068B (en) * | 2019-04-08 | 2021-07-09 | 东南大学 | Large-scale coal-fired power plant CO based on neural network inverse control2Trapping system and feedforward control method |
| CN110026068A (en) * | 2019-04-08 | 2019-07-19 | 东南大学 | A kind of large-scale coal fired power plant CO based on Neural network inverse control2Trapping system and feed forward control method |
| CN109876658A (en) * | 2019-04-08 | 2019-06-14 | 国电环境保护研究院有限公司 | Subregion vortex volume inhales spray ammonia system and spray ammonia adjusting method |
| CN110299188A (en) * | 2019-05-09 | 2019-10-01 | 上海电力学院 | SCR flue gas denitrification system GRNN modeling method based on GA variables choice |
| CN110908351A (en) * | 2019-11-25 | 2020-03-24 | 东南大学 | Support vector machine-fused SCR denitration system disturbance suppression prediction control method |
| CN110908351B (en) * | 2019-11-25 | 2022-11-18 | 东南大学 | Support vector machine-fused SCR denitration system disturbance suppression prediction control method |
| CN112488145A (en) * | 2019-11-26 | 2021-03-12 | 大唐环境产业集团股份有限公司 | NO based on intelligent methodxOnline prediction method and system |
| CN112488145B (en) * | 2019-11-26 | 2024-05-28 | 大唐环境产业集团股份有限公司 | NO based on intelligent methodxOnline prediction method and system |
| CN110935312A (en) * | 2019-12-16 | 2020-03-31 | 广州珠江电力有限公司 | Dynamic monitoring device and dynamic monitoring method for SCR flue gas denitration system |
| CN111413938A (en) * | 2020-04-16 | 2020-07-14 | 南京英璞瑞自动化科技有限公司 | SCR denitration system disturbance suppression prediction control method based on converted ammonia injection amount |
| CN111589301A (en) * | 2020-05-29 | 2020-08-28 | 广东电科院能源技术有限责任公司 | Method, device, equipment and storage medium for predicting SCR denitration performance of coal-fired power plant |
| CN111897373A (en) * | 2020-08-05 | 2020-11-06 | 海南创实科技有限公司 | Model prediction-based ammonia injection flow adjusting method for SCR denitration device |
| CN111897373B (en) * | 2020-08-05 | 2022-11-01 | 海南创实科技有限公司 | Model prediction-based ammonia injection flow adjusting method for SCR denitration device |
| CN112180731B (en) * | 2020-10-13 | 2024-05-31 | 天津大学 | Energy equipment operation control method and system |
| CN112180731A (en) * | 2020-10-13 | 2021-01-05 | 天津大学 | Energy equipment operation control method and system |
| CN112461995A (en) * | 2020-11-03 | 2021-03-09 | 西安热工研究院有限公司 | Method for predicting ammonia escape of thermal power plant |
| CN112596474A (en) * | 2020-11-23 | 2021-04-02 | 山东微立方信息技术股份有限公司 | Uniformity optimization method for SCR denitration system |
| CN112651166A (en) * | 2020-11-24 | 2021-04-13 | 呼和浩特科林热电有限责任公司 | Denitration system inlet nitrogen oxide concentration prediction method and device and denitration system |
| CN112619394A (en) * | 2020-11-24 | 2021-04-09 | 呼和浩特科林热电有限责任公司 | Denitration ammonia injection self-adaptive control method and device and denitration system |
| CN112627953A (en) * | 2020-12-22 | 2021-04-09 | 上海海事大学 | Ship SCR system ammonia injection amount control method based on ARMAX and MMPC |
| CN112933913A (en) * | 2021-02-08 | 2021-06-11 | 国家能源集团国源电力有限公司 | Ammonia injection control method and device and coal combustion system |
| CN113144843A (en) * | 2021-03-16 | 2021-07-23 | 国家能源集团国源电力有限公司 | Ammonia spraying control method and device |
| CN113488111B (en) * | 2021-07-05 | 2022-11-22 | 东南大学 | Ammonia injection amount optimization modeling method for SCR denitration system |
| CN113488111A (en) * | 2021-07-05 | 2021-10-08 | 东南大学 | Ammonia injection amount optimization modeling method for SCR denitration system |
| CN113398758A (en) * | 2021-07-20 | 2021-09-17 | 山信软件股份有限公司 | Ammonia water spraying flow control system and control method thereof |
| CN113996158A (en) * | 2021-08-16 | 2022-02-01 | 国能(山东)能源环境有限公司 | Control method and device for external desulfurization, electronic equipment and storage medium |
| CN114326375A (en) * | 2021-12-22 | 2022-04-12 | 江苏精瓷智能传感技术研究院有限公司 | Nitrogen oxygen sensor pump unit self-adaptive control system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107694337A (en) | Coal unit SCR denitrating flue gas control methods based on network response surface | |
| Catalao et al. | An artificial neural network approach for short-term wind power forecasting in Portugal | |
| CN106022521B (en) | Short-term load prediction method of distributed BP neural network based on Hadoop architecture | |
| CN106599586B (en) | SCR neural network based intelligently sprays ammonia optimization method and device | |
| CN108764568B (en) | Data prediction model tuning method and device based on LSTM network | |
| CN107967803A (en) | Traffic congestion Forecasting Methodology based on multi-source data and variable-weight combined forecasting model | |
| CN109214579B (en) | BP neural network-based saline-alkali soil stability prediction method and system | |
| RU2013130996A (en) | OPTIMIZED INTEGRATED CONTROL FOR A POWER PLANT OPERATING OXYGEN FUEL COMBUSTION | |
| CN105138717A (en) | Transformer state evaluation method by optimizing neural network with dynamic mutation particle swarm | |
| CN104408518A (en) | Method of learning and optimizing neural network based on particle swarm optimization algorithm | |
| CN103778482A (en) | Aquaculture dissolved oxygen short-term prediction method based on multi-scale analysis | |
| CN113031555B (en) | Intelligent purification system for harmful gas in environment of livestock and poultry house | |
| CN106251242B (en) | Wind power output interval combination prediction method | |
| CN106022471A (en) | Wavelet neural network model ship rolling real-time prediction method based on particle swarm optimization algorithm | |
| CN107194026A (en) | Absorption tower sweetening process modeling method based on Bayesian network | |
| KR20210019311A (en) | Method and device for forecasting flood based on data analyzing | |
| CN107145935A (en) | A kind of Smith Prediction Control methods based on modified neutral net | |
| CN108615097A (en) | A kind of wind speed forecasting method, system, equipment and computer readable storage medium | |
| CN112488145A (en) | NO based on intelligent methodxOnline prediction method and system | |
| CN108717506A (en) | A method of prediction coke hot strength | |
| CN110059821A (en) | Neural network optimization, device, computer equipment and storage medium | |
| CN107703760A (en) | Absorbing natural gas tower sweetening process control method based on RBF and GDHP | |
| Ghanbari et al. | Prediction of degree of crystallinity for the LTA zeolite using artificial neural networks | |
| CN116700003A (en) | Method and system for constructing reinforcement learning environment by using process industry historical data | |
| CN115034159A (en) | Power prediction method, device, storage medium and system for offshore wind farm |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180216 |
|
| WD01 | Invention patent application deemed withdrawn after publication |