CN109992844A - A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model - Google Patents
A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model Download PDFInfo
- Publication number
- CN109992844A CN109992844A CN201910189111.4A CN201910189111A CN109992844A CN 109992844 A CN109992844 A CN 109992844A CN 201910189111 A CN201910189111 A CN 201910189111A CN 109992844 A CN109992844 A CN 109992844A
- Authority
- CN
- China
- Prior art keywords
- adqpso
- svr
- carbon content
- particle
- svr model
- 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
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 title claims abstract description 45
- 229910052799 carbon Inorganic materials 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 34
- 239000010881 fly ash Substances 0.000 title claims abstract description 29
- 230000003044 adaptive effect Effects 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 6
- 239000002245 particle Substances 0.000 claims description 69
- 238000005457 optimization Methods 0.000 claims description 24
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 239000003546 flue gas Substances 0.000 claims description 6
- 239000003245 coal Substances 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims description 3
- 229910052760 oxygen Inorganic materials 0.000 claims description 3
- 239000001301 oxygen Substances 0.000 claims description 3
- 239000000446 fuel Substances 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 44
- 238000010276 construction Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 32
- 239000003500 flue dust Substances 0.000 description 16
- 230000000694 effects Effects 0.000 description 7
- 230000002068 genetic effect Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 238000010248 power generation Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000004744 fabric Substances 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 229910052698 phosphorus Inorganic materials 0.000 description 2
- 239000011574 phosphorus Substances 0.000 description 2
- 230000005619 thermoelectricity Effects 0.000 description 2
- 238000010977 unit operation Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 238000000342 Monte Carlo simulation Methods 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 235000019504 cigarettes Nutrition 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005243 fluidization Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of boiler flyash carbon content prediction techniques based on ADQPSO-SVR model, this method carries out the optimizing of SVR parameter using ADQPSO algorithm, adaptive precocious decision criteria and hybrid perturbation operation is added, optimal parameter combination is found out, carries out model construction as training sample using presence feature data.Compared with prior art, boiler flyash carbon content prediction technique accuracy and generalization ability based on ADQPSO-SVR model are more preferable, have great importance to thermal power plant.
Description
Technical field
The present invention relates to a kind of boiler flyash carbon content prediction techniques, are based on ADQPSO-SVR model more particularly, to one kind
Boiler flyash carbon content prediction technique.
Background technique
The first half of the year in 2018, thermoelectricity generated energy increase by 8.0% on a year-on-year basis, and thermoelectricity is promoted 116 hours on year-on-year basis using hour.China
Power supply architecture is based on thermal power generation, and wherein coal fired power generation occupies leading position in thermal power generation.
In face of the double constraints of resource and environment, the form that Thermal Power Generation Industry faces is more and more severeer, it means that firepower hair
The transition and upgrade of power technology is extremely urgent, it is necessary into the developing stage of clean and effective, just adapt to the development in epoch and society.
Unburned carbon in flue dust is the important parameter closely related with boiler efficiency, therefore, is only accurately surveyed to unburned carbon in flue dust
Amount just can guarantee the efficiency of boiler.The method of on-line monitoring unburned carbon in flue dust has much at present, mainly there is microwave absorption method, fluidisation
Bed CO2 mensuration.But the disadvantages such as anti-interference ability is weak, precision is low that the above method has.Therefore many researchers have also been proposed
Unburned carbon in flue dust monitoring model is established based on the relatively advanced measurement method such as genetic neural network and support vector machines, but on
The disadvantages of it is not strong that the measurement method stated can have a generalization ability, easily falls into Local Minimum, so, under this kind of background, how
Finding corresponding method for precisely solving becomes emphasis and difficulties in current industry and academia.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide one kind to be based on ADQPSO-
The boiler flyash carbon content prediction technique of SVR model.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model, method includes the following steps:
Step S1: setting quantum particle swarm initial value;
Step S2: assessment quantum particle swarm records the fitness of each particle, the personal best particle of particle, population
Global optimum position and average optimal location;
Step S3: particle position is updated, the fitness of each particle, the personal best particle of particle, particle are recalculated
Group global optimum position and average optimal location;
Step S4: judging whether quanta particle swarm optimization termination condition meets, if so, step S6 is executed, if it is not, executing step
Rapid S5;
Step S5: calculating the fitness variance of group, adaptive precocious judgement is carried out, if fitness variance is less than threshold
Value executes disturbance operation to group, then executes step S3, otherwise directly execute step S3;
Step S6: output population global optimum position is as SVR optimal value of the parameter;
Step S7: historical data is acquired from the automation equipment of scene operation, training sample is obtained from historical data
Input variable;
Step S8: SVR model is constructed using the SVR optimal value of the parameter that step S6 is obtained;
Step S9: in the SVR model that sample point input step S8 is obtained, boiler flyash carbon content is obtained.
The historical data includes boiler load, coal-supplying amount, economizer exit flue-gas temperature, the pivot angle of burner, combustion
Expect windshield plate aperture, burnout degree baffle opening, primary wind pressure, secondary wind pressure, flue gas oxygen content, burner hearth and bellows differential pressure and coal
Kind characteristic.
The SVR model uses Gauss Radial basis kernel function.
The sample point in input variable and step S9 in the step S7 obtains normalized value by pretreatment, described
SVR model output boiler flyash carbon content normalized value, normalized value inverse transformation is then obtained into boiler flyash carbon content
Data.
To the data that can not directly measure on site, the historical data is according to SIS system, automation equipment in power station
Or traditional performance evaluation obtains estimated value.
The function of the fitness is mean square error function.
The disturbance operation specifically: use hybrid perturbation operator, increase in average optimal position and disturb, described is mixed
Close the operator and the operator for obeying Cauchy's distribution that disturbing operator includes Gaussian distributed.
The hybrid perturbation operator are as follows:
βk=e1Ck(0,1)+e2Nk(0,1)
In formula: βkFor hybrid perturbation operator, k is current iteration number, Ck(0,1) and NkIt (0,1) is respectively to obey Cauchy point
The random number of cloth and Gaussian Profile, e1,e2For coefficient of disturbance, expression formula is as follows:
In formula: e1maxAnd e1minRespectively e1Maximum and minimum value;e2maxAnd e2minRespectively e2Dominant bit and minimum
Value;kmaxFor maximum number of iterations.
Compared with prior art, the invention has the following advantages that
(1) SVR is optimized using ADQPSO algorithm, it is quasi- that adaptive precocious judgement is added in quantum telepotation
It is then operated with hybrid perturbation, finds out optimal parameter combination, than conventional boiler flyash carbon content prediction model accuracy and general
Change ability is more preferable, has great importance to thermal power plant.
(2) SVR uses Gauss Radial basis kernel function, and calculation amount is few, wide adaptation range.
(3) parameter under each load condition is contained in the historical data acquired, represents each operating condition of unit operation
State, have it is comprehensive.
(4) fitness function selects mean square error, and error between prediction result and legitimate reading can be made minimum.
Detailed description of the invention
Fig. 1 is flow chart of the invention;
Fig. 2 is the prediction effect curve of GA-SVR, PSO-SVR, QPSO-SVR and ADQPSO-SVR model unburned carbon in flue dust
Schematic diagram;
Fig. 3 is that the prediction effect of GA-SVR, PSO-SVR, QPSO-SVR and ADQPSO-SVR model unburned carbon in flue dust is opposite
Percentage error schematic diagram;
Fig. 4 is the prediction effect curve synoptic diagram of the unburned carbon in flue dust of GA-BP, GA-RBF and ADQPSO-SVR model;
Fig. 5 is that the prediction effect percentage ratio error of GA-BP, GA-RBF and ADQPSO-SVR model unburned carbon in flue dust is shown
It is intended to.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention
Premised on implemented, the detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to
Following embodiments.
Embodiment
A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model, method includes the following steps:
Step S1: setting quantum particle swarm initial value;
Step S2: assessment quantum particle swarm records the fitness of each particle, the personal best particle of particle, population
Global optimum position and average optimal location;
Step S3: particle position is updated, the fitness of each particle, the personal best particle of particle, particle are recalculated
Group global optimum position and average optimal location;
Step S4: judging whether quanta particle swarm optimization termination condition meets, if so, step S6 is executed, if it is not, executing step
Rapid S5;
Step S5: calculating the fitness variance of group, adaptive precocious judgement is carried out, if fitness variance is less than threshold
Value executes disturbance operation to group, then executes step S3, otherwise directly execute step S3;
Step S6: output population global optimum position is as SVR optimal value of the parameter;
Step S7: historical data is acquired from the automation equipment of scene operation, training sample is obtained from historical data
Input variable;
Step S8: SVR model is constructed using the SVR optimal value of the parameter that step S6 is obtained;
Step S9: in the SVR model that sample point input step S8 is obtained, boiler flyash carbon content is obtained.
Historical data includes boiler load, coal-supplying amount, economizer exit flue-gas temperature, the pivot angle of burner, fuel windscreen
Plate aperture, burnout degree baffle opening, primary wind pressure, secondary wind pressure, flue gas oxygen content, burner hearth and bellows differential pressure and coal are special
Property.
SVR model uses Gauss Radial basis kernel function.
The sample point in input variable and step S9 in step S7 obtains normalized value by pretreatment, and SVR model is defeated
Then normalized value inverse transformation is obtained the data of boiler flyash carbon content by the normalized value of boiler flyash carbon content out.
To the data that can not directly measure on site, historical data is according to safety instrument (Safety in power station
Instrumented System, SIS) system, automation equipment or traditional performance evaluation obtain estimated value.
The function of fitness is mean square error function.
Disturbance operation specifically: hybrid perturbation operator is used, increases in average optimal position and disturbs, hybrid perturbation operator packet
It includes the operator of Gaussian distributed and obeys the operator of Cauchy's distribution.
Hybrid perturbation operator are as follows:
βk=e1Ck(0,1)+e2Nk(0,1)
In formula: βkFor hybrid perturbation operator, k is current iteration number, Ck(0,1) and NkIt (0,1) is respectively to obey Cauchy point
The random number of cloth and Gaussian Profile, e1,e2For coefficient of disturbance, expression formula is as follows:
In formula: e1maxAnd e1minRespectively e1Maximum and minimum value;e2maxAnd e2minRespectively e2Dominant bit and minimum
Value;kmaxFor maximum number of iterations.
Adaptive disturbance quanta particle swarm optimization (Adaptive Disturbance Quantum-behaved
Particle Swarm Optimization, ADQPSO) it is a kind of optimizing algorithm, support vector regression (Support Vector
Regression, SVR) it is a kind of sorting algorithm, the present invention obtains SVR optimized parameter using ADQPSO algorithm, then establishes
ADQPSO-SVR model.
ADQPSO-SVR model based on regression algorithm building is for handling regression problem, and the mathematical description of regression problem is such as
Under:
If training set is { (x1,y1),(x2,y2)···,(xn,yn), wherein input vector xi∈Rn, output vector yi
∈Rn.Construct regression function:
F (x)=(wx)+b (1)
W is weight vector in formula (1), and b is threshold value, is re-introduced into error function.In the present invention, Huber loss is introduced
Function, and it is considered as the error function in model.It is now assumed that the spacing between sample and regression function is below insensitive loss
Regression problem can be then converted to secondary convex optimization problem by variable ε, detailed to be described as follows:
Wherein, l is number of samples, in actual environment, often contains more noise, therefore need to introduce two relaxations
The factor, i.e. ξiAnd ξi *, then formula (2) and formula (3) are arranged as follows:
In formula (4), c is penalty factor.ξiIt indicates to be located at the sample point above regression function, ξi *It indicates to be located at regression function
The sample point of lower section.
Lagrangian Arithmetic processing is introduced to formula (4) and formula (5), at extreme point position, to w, b, ξiAnd ξi *Deng
Variable does local derviation processing, and local derviation value is 0, and formula (6) then can be obtained:
It can be obtained according to formula (6) translation transformation:
It is solved formula (7) to obtain ξiAnd ξi *, then by KKT condition, can obtain b, b is then brought into regression equation
In, obtain following optimal equation of linear regression:
WithRemove replacement xi;Substantially, inner product processing as Mercer core, expression formula such as formula (9):
Based on analysis and reasoning process as above, corresponding recurrence processing function under nonlinear condition is obtained:
It is not difficult to obtain the main algorithm process of SVR from above discussion and reasoning:, can will be original by kernel function
Input data, which projects to, can divide and in linear another dimension group, and problem to be processed is then switched to linear problem;And in the calculation
In method, the most key is the kernel function that design and selection match with problem to be processed, i.e. the selection direct relation of kernel function
To projection array structure, determine indirectly SVR performance and in terms of the characteristics of.Currently, in SVR the most often
Kernel function mainly has following three types:
(a) Polynomial kernel function:
k(x,xi)=[(xxi)+1]q (11)
Wherein, q is polynomial order.
(b) Sigmoid kernel function:
k(x,xi)=tanh [v (xxi)+c0] (12)
Wherein, v is single order constant, c0For bias term.
(c) Gauss Radial basis kernel function:
k(x,xi)=exp [- | x-xi|2/σ2] (13)
Wherein, σ is radial base core coefficient variation.
For the above three classes kernel function, highest dimension is (a) class, and the error amount of the function is also bigger, so it is counted
Calculation amount is bigger;And for (b) class kernel function, c0Bigger with limitation of the v in terms of value, only component values meet
Mercer condition so only certain numerical value can guarantee that the function is positive definite, therefore exists centainly in terms of adaptation range
Limitation;Different from preceding two class function, (c) application of the class function in SVR is most commonly used.Therefore, the present invention uses
Gauss Radial basis kernel function.
Particle swarm optimization algorithm (PSO) is to be based on group's intelligence in one kind that nineteen ninety-five proposes by Kennedy and Eberhart
The global optimization approach of energy.The system has a particle group, and each particle represents the possible solution of optimization problem.There are
M particle, in the original PSO algorithm of D dimension space, position vector of m-th of particle in kth time iteration step: Xm(k)=(Xm1
(k),Xm2(k),···,XmD(k)), velocity vector: Vm(k)=(Vm1(k),Vm2(k),···,VmD(k)).Particle according to
Following iterative formula updates position and speed are as follows:
Xm(k+1)=Xm1(k)+Vd(k+1) (14)
Vd(k+1)=ω Vm(k)+c1r1(Pm(k)-Xm(k))+c2r2(Pg(k)-Xm(k)) (15)
Wherein, m=1,2 ..., M, d=1,2 ..., D;ω is inertia weight, c1,c2Indicate accelerator coefficient;r1,r2It is uniform
The random number being distributed in (0,1);The optimum position of i-th of particle is by vector Pm=(Pm1,Pm2,···,PmD) indicate;It is whole
The global optimum position of a population is by vector Pg=(Pg1,Pg2,···,PgD) indicate.But particle swarm optimization algorithm cannot
Guarantee to search globally optimal solution with probability 1, there are certain defects for the global convergence of PSO.In view of the above-mentioned problems, Sun is from amount
The angle of sub- mechanics investigates the evolution modelling of particle individual and group, proposes QPSO algorithm.Particle band in QPSO algorithm
There is location information, without velocity information, the control parameter of algorithm is relatively less, compared with PSO, with global optimizing ability
By force, robustness is good and calculates time-consuming short advantage.
Also assume that particle m is moved in D dimension space, at the kth iteration, the potential well that particle m is tieed up in D are as follows:
pmD(k)=r1PmD(k)+[1-r2]pgD(k) (16)
Using the available L-expression that particle m is tieed up in D in+1 iteration of kth of Monte Carlo method are as follows:
xmD(k+1)=PmD(k)±0.5LmD(k)ln[1/r] (17)
In formula: r is generally evenly distributed in the random number on (0,1);LmD(k) it can be determined by following formula.
LmD(k)=2 β | sd(k)-xmd(k)| (18)
In formula, sd(k) particle average optimal position is indicated;β is dynamic expanding-constriction coefficient, generally takes 0.5-1.sd(k)
It is the central point of all particle personal best particles, can be calculated by following formula:
In formula, PiDifferent particle personal best particles are represented, composite type (17), (18) obtain the position of quantum particle swarm more
New equation:
xiD(k+1)=piD(k)±β|sd(k)-xiD(k)|·ln(1/r) (20)
In formula, sdIt (k) is particle average optimal position, unlike particle swarm algorithm, quanta particle swarm optimization is not required to
Velocity vector is wanted, algorithm is made to be easier to execute.In order to solve premature problem, joined in quanta particle swarm optimization adaptive
Precocious judgement, disturbing operator and dynamic expanding constriction coefficient are answered, ADQPSO algorithm is formd.In QPSO algorithm, according to suitable
The state of response changed to judge group.
The function of the fitness of group are as follows:
In formula: f is the echo cancellation factor, fmThe fitness of particle m, favgFor particle group average fitness.
As group fitness s2When less than threshold value, to avoid precocious phenomenon, increases disturbance operation and algorithm is allowed to jump out part most
Excellent to seek solution, the other positions for reaching space scan for again.Common disturbance operation, which mainly increases in global optimum position, disturbs
It is dynamic, increase disturbance in average optimal position and all increase by 3 kinds of disturbance in global optimum position and average optimal location and operate.This hair
It is bright to be operated based on basic Gaussian Profile and Cauchy's distribution using hybrid perturbation, increase disturbance behaviour in average optimal position
Make.Disturb the operator expression formula of operation are as follows:
βk=e1Ck(0,1)+e2Nk(0,1) (22)
In formula: k is current iteration number, Ck(0,1) and Nk(0,1) be respectively obey Cauchy distribution and Gaussian Profile with
Machine number.e1,e2For coefficient of disturbance.Expression formula is as follows:
In formula: e1maxAnd e1minRespectively e1Maximum and minimum value;e2maxAnd e2minRespectively e2Dominant bit and minimum
Value;kmaxFor maximum number of iterations.Average optimal position after being operated using hybrid perturbationAre as follows:
Hybrid perturbation operator considers Gauss number and Cauchy's random number, meanwhile, introduce coefficient of disturbance e1And e2, when
e2When value is larger, hybrid perturbation operator approximation Gaussian distributed;Work as e2When value is smaller, hybrid perturbation operator is then similar to
Cauchy's distribution is obeyed, to obtain random number in extensive range.Meanwhile lesser e1Convenient for the region of search to current particle into
Row part precise search, is conducive to algorithmic statement;Biggish e1Be conducive to that algorithm is promoted to jump out local minimum point, the enhancing overall situation is searched
Suo Nengli.
Particle swarm intelligence algorithm is introduced into research by the present invention, is constructed the pattern of fusion based on support vector machines and is returned mould
Adaptive precocious decision criteria and hybrid perturbation operation is added to support vector machines parameter in type in quantum particle swarm optimization
It optimizes.Substantially, this kind of algorithm makes optimization processing mainly for the parameter in SVR, and then constructs optimal prediction model.
In SVR, precision of prediction is mainly by ε (insensitive loss variable), σ (radial base core coefficient variation) and c (punishment parameter)
Deng decision.ε variable is used to control the width in non-sensitive area in sample, and specific numerical value is related to the operation in sample;And c becomes
The specific value size of amount is used to indicate punishment situation suffered by sample point except ε pipeline, and value can more significantly shadow
Performance in operation is rung to optimization ADQPSO-SVR model;σ is used to indicate the correlation circumstance of each parameter in SVR.SVR ensemble stream
Journey are as follows:
In order to guarantee the forecasting accuracy of model, by 150 groups of data of automation equipment system acquisition of power plant as instruction
Practice sample, simultaneously as usually there is the case where operating condition variation during operation in Power Plant, therefore the automation equipment number acquired
According to the state for each operating condition that must can represent unit operation, therefore acquire the ginseng contained under each load condition in data
Number.Partial data is as shown in table 1.
The partial data of 1 unburned carbon in flue dust of table
For the present invention using MATLAB as experiment porch, computer hardware is configured to 2.4GHz CPU, 8GB memory, operating system
It is trained to obtain final learning parameter search range are as follows: penalty factor c is between 0-1000, no for 64 Windows 10
Susceptibility ε is between 0.001-1, and radial base nuclear parameter σ is between 0.01-1.50 groups of flying dusts are contained using ADQPSO-SVR model
Carbon amounts data are predicted, and particle swarm algorithm (PSO), genetic algorithm (GA), quanta particle swarm optimization (QPSO) is respectively adopted
Optimizing, and relatively more constructed ADQPSO- are carried out with learning parameter of adaptive disturbance quanta particle swarm optimization (ADQPSO) to SVR
The prediction effect of SVR model.
The fitness and learning parameter of each algorithm of table 2
Above-mentioned 4 kinds of algorithm parameters: population size 50, maximum number of iterations 200;The study of particle swarm algorithm (PSO)
The factor is 2, weight factor 1;The β of quanta particle swarm optimization (QPSO) takes 0.7;Threshold value takes 3 × 10-6;In order to avoid each algorithm
Randomness, above 4 kinds of algorithms are separately operable 20 times, fitness function select mean square error (MSE), obtain 4 kinds of algorithms
Training result and final optimizing parameter are as shown in table 2.
For table in analysis it is found that the average fitness of ADQPSO algorithm is minimum, the average fitness of Genetic Algorithms is maximum, says
The optimizing ability of bright ADQPSO algorithm is most strong, and the optimizing ability of Genetic Algorithms is most weak.Meanwhile Genetic Algorithms and PSO algorithm
Average fitness it is all bigger than the average fitness of QPSO algorithm, illustrate QPSO algorithm aspect of performance than Genetic Algorithms and
PSO algorithm is good, therefore adaptively disturbance quanta particle swarm optimization (ADQPSO) can be SVR selection optimal parameter, therefore ADQPSO-
SVR model is more preferable than the ADQPSO-SVR forecast result of model of tradition SVR and other parameters optimization, more conducively boiler fly ash
The prediction of phosphorus content.
50 groups of unburned carbon in flue dust data are predicted using ADQPSO algorithm optimization parameter building ADQPSO-SVR model,
Predict that the evaluation index of error selects mean absolute percentage error (MAPE), percentage ratio error (RPE) and root mean square to miss
Poor (RMSE).The consensus forecast effect curve of unburned carbon in flue dust is as shown in Fig. 2, unburned carbon in flue dust evaluation index is as shown in table 3.4
The PRE distribution situation of kind model is as shown in table 4 and Fig. 3.
The MAPE and RMSE of 34 kinds of model predictions of table
The RPE distribution situation of 44 kinds of model predictions of table
By the above chart it is found that the RPE of the unburned carbon in flue dust prediction model of ADQPSO-SVR has 32 points in 5%-15%
Between, account for about the 64% of total amount.< 5% number is 7, and the number of GA-SVR model and PSO-SVR model < 5% is 1,
The number of QPSO-SVR model < 5% is 3.ADQPSO-SVR prediction model is better than other algorithm models on the whole, has highest
Precision of prediction.In order to further verify model built in the application of unburned carbon in flue dust, the more common GA-BP and GA-RBF of use
Algorithm is compared with this prediction model.And in order to verify the generalization ability of this model, 50 groups of data is optionally taken to carry out moulds at random
Type prediction, for the purposes of avoiding randomness, 3 kinds of models are separately operable 20 times, and average prediction curve is as shown in Figure 4:
The evaluation index of 3 kinds of models is as shown in table 5, it is seen that ADQPSO-SVR prediction model relative error remains to control well
System illustrates that the model has good generalization ability in accuracy rating.The RPE distribution situation of 3 kinds of models such as table 6 and Fig. 5 institute
Show.
The MAPE and RMSE of 53 kinds of model predictions of table
The RPE distribution situation of 63 kinds of model predictions of table
In ADQPSO-SVR prediction model RPE, there is 72% point in 5%-15%, ratio is higher than GA-BP model and GA-
RBF model, and in ADQPSO-SVR model RPE, the point greater than 20% only has 1, and GA-BP model has 6, GA-RBF model
There are 7.Therefore, requirement, prediction effect can be reached in precision by establishing the boiler flyash carbon content model based on ADQPSO-SVR
Better than the SVR of traditional SVR and other parameters optimization, and generalization ability is also relatively strong, compared with GA-BP model and GA-RBF model
Also there is higher precision of prediction.
The forecasting problem for flying cigarette phosphorus content for boiler, this paper presents ADQPSO-SVR prediction model, this method is being measured
It joined adaptive precocious decision criteria and hybrid perturbation operation on the basis of sub- particle group optimizing, by with other common 3
The algorithm of kind optimization SVR parameter is compared, and the improvement of ADQPSO-SVR model optimizes the problem of learning parameter selection in SVR, is better than
The SVR of traditional SVR and other parameters optimization, by compared with GA-BP and GA-RBF algorithm, proposed modeling method
Computational accuracy, prediction error and in terms of compare with application advantage, by modeling and simulation the result shows that, herein
The unburned carbon in flue dust prediction model precision of prediction with higher and generalization ability of the SVR used has good popularization and application
Prospect.It, need to be to the lag time of auxiliary variable each in modeling process in order to overcome disturbing factor to measure in development from now on
Accurately estimated, such prediction model can make the value of unburned carbon in flue dust more accurate.
Claims (8)
1. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model, which is characterized in that this method include with
Lower step:
Step S1: setting quantum particle swarm initial value;
Step S2: it is global to record the fitness of each particle, the personal best particle of particle, population for assessment quantum particle swarm
Optimal location and average optimal location;
Step S3: particle position is updated, it is complete to recalculate the fitness of each particle, the personal best particle of particle, population
Office's optimal location and average optimal location;
Step S4: judging whether quanta particle swarm optimization termination condition meets, if so, step S6 is executed, if it is not, executing step
S5;
Step S5: calculating the fitness variance of group, carries out adaptive precocious judgement, right if fitness variance is less than threshold value
Group executes disturbance operation, then executes step S3, otherwise directly executes step S3;
Step S6: output population global optimum position is as SVR optimal value of the parameter;
Step S7: historical data is acquired from the automation equipment of scene operation, the input of training sample is obtained from historical data
Variable;
Step S8: SVR model is constructed using the SVR optimal value of the parameter that step S6 is obtained;
Step S9: in the SVR model that sample point input step S8 is obtained, boiler flyash carbon content is obtained.
2. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model according to claim 1, special
Sign is that the historical data includes boiler load, coal-supplying amount, economizer exit flue-gas temperature, the pivot angle of burner, fuel
Windshield plate aperture, burnout degree baffle opening, primary wind pressure, secondary wind pressure, flue gas oxygen content, burner hearth and bellows differential pressure and coal
Characteristic.
3. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model according to claim 1, special
Sign is that the kernel function of the SVR model is Gauss Radial basis kernel function.
4. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model according to claim 1, special
Sign is that the input variable in the step S7 and the sample point in step S9 are described by pretreatment acquisition normalized value
SVR model output boiler flyash carbon content normalized value, normalized value inverse transformation is obtained into boiler flyash carbon content.
5. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model according to claim 1, special
Sign is that the historical data that scene can not directly measure if it exists is then passed using SIS system or automation equipment in power station
System performance evaluation obtains estimated value, as historical data.
6. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model according to claim 1, special
Sign is that the function of the fitness is mean square error function.
7. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model according to claim 1, special
Sign is that the disturbance operates specifically: uses hybrid perturbation operator, increases in average optimal position and disturb, described is mixed
Close the operator and the operator for obeying Cauchy's distribution that disturbing operator includes Gaussian distributed.
8. a kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model according to claim 7, special
Sign is, the hybrid perturbation operator are as follows:
βk=e1Ck(0,1)+e2Nk(0,1)
In formula: βkFor hybrid perturbation operator, k is current iteration number, Ck(0,1) and Nk(0,1) be respectively obey Cauchy distribution and
The random number of Gaussian Profile, e1、e2For coefficient of disturbance, e1、e2Expression formula is as follows:
In formula: e1maxAnd e1minRespectively e1Maximum and minimum value, e2maxAnd e2minRespectively e2Dominant bit and minimum value,
kmaxFor maximum number of iterations.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910189111.4A CN109992844A (en) | 2019-03-13 | 2019-03-13 | A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910189111.4A CN109992844A (en) | 2019-03-13 | 2019-03-13 | A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109992844A true CN109992844A (en) | 2019-07-09 |
Family
ID=67130557
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910189111.4A Pending CN109992844A (en) | 2019-03-13 | 2019-03-13 | A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109992844A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110852480A (en) * | 2019-10-14 | 2020-02-28 | 杭州中恒云能源互联网技术有限公司 | Electric power data completion method |
CN111189638A (en) * | 2019-12-24 | 2020-05-22 | 沈阳化工大学 | HMM and QPSO optimization algorithm-based bearing fault degree identification method |
CN111754093A (en) * | 2020-06-11 | 2020-10-09 | 浙江浙能技术研究院有限公司 | Fly ash carbon content prediction method based on coal quality prediction and PSO-SVM |
CN114637212A (en) * | 2022-05-17 | 2022-06-17 | 浙江浩普智能科技有限公司 | Oxygen content prediction control method and system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102778538A (en) * | 2012-07-06 | 2012-11-14 | 广东电网公司电力科学研究院 | Soft measuring method based on improved SVM (Support Vector Machine) for measuring boiler unburned carbon content in fly ash |
CN106649919A (en) * | 2016-09-13 | 2017-05-10 | 上海电气电站环保工程有限公司 | Method and system for predicting carbon content of fly ash in coal-fired power plant boiler |
CN107133460A (en) * | 2017-04-26 | 2017-09-05 | 中国能源建设集团广东省电力设计研究院有限公司 | A kind of online dynamic prediction method of boiler flyash carbon content |
-
2019
- 2019-03-13 CN CN201910189111.4A patent/CN109992844A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102778538A (en) * | 2012-07-06 | 2012-11-14 | 广东电网公司电力科学研究院 | Soft measuring method based on improved SVM (Support Vector Machine) for measuring boiler unburned carbon content in fly ash |
CN106649919A (en) * | 2016-09-13 | 2017-05-10 | 上海电气电站环保工程有限公司 | Method and system for predicting carbon content of fly ash in coal-fired power plant boiler |
CN107133460A (en) * | 2017-04-26 | 2017-09-05 | 中国能源建设集团广东省电力设计研究院有限公司 | A kind of online dynamic prediction method of boiler flyash carbon content |
Non-Patent Citations (1)
Title |
---|
陈道君等: ""基于自适应扰动量子粒子群算法参数优化的支持向量回归机短期风电功率预测"", 《电网技术》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110852480A (en) * | 2019-10-14 | 2020-02-28 | 杭州中恒云能源互联网技术有限公司 | Electric power data completion method |
CN110852480B (en) * | 2019-10-14 | 2023-03-10 | 杭州中恒云能源互联网技术有限公司 | Electric power data completion method |
CN111189638A (en) * | 2019-12-24 | 2020-05-22 | 沈阳化工大学 | HMM and QPSO optimization algorithm-based bearing fault degree identification method |
CN111189638B (en) * | 2019-12-24 | 2021-08-06 | 沈阳化工大学 | HMM and QPSO optimization algorithm-based bearing fault degree identification method |
CN111754093A (en) * | 2020-06-11 | 2020-10-09 | 浙江浙能技术研究院有限公司 | Fly ash carbon content prediction method based on coal quality prediction and PSO-SVM |
CN111754093B (en) * | 2020-06-11 | 2023-09-22 | 浙江浙能技术研究院有限公司 | Fly ash carbon content prediction method based on coal quality prediction and PSO-SVM |
CN114637212A (en) * | 2022-05-17 | 2022-06-17 | 浙江浩普智能科技有限公司 | Oxygen content prediction control method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109992844A (en) | A kind of boiler flyash carbon content prediction technique based on ADQPSO-SVR model | |
Huang et al. | Air quality prediction using improved PSO-BP neural network | |
CN107016455B (en) | Prediction system and method for oxygen content of flue gas at hearth outlet of circulating fluidized bed household garbage incineration boiler | |
CN109583585B (en) | Construction method of power station boiler wall temperature prediction neural network model | |
Wang et al. | Forecasting Chinese provincial carbon emissions using a novel grey prediction model considering spatial correlation | |
CN106224939B (en) | Circulating fluid bed domestic garbage burning boiler bed temperature Forecasting Methodology and system | |
CN105864797B (en) | Real-time prediction system and method for boiler entering heat value of circulating fluidized bed household garbage incineration boiler | |
CN112085277B (en) | SCR denitration system prediction model optimization method based on machine learning | |
CN107133460A (en) | A kind of online dynamic prediction method of boiler flyash carbon content | |
CN106649919A (en) | Method and system for predicting carbon content of fly ash in coal-fired power plant boiler | |
CN103728879A (en) | Power station boiler emission soft measuring method based on least squares support vector machine and on-line updating | |
CN112613237B (en) | CFB unit NOx emission concentration prediction method based on LSTM | |
CN111814956A (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
CN106327004A (en) | Cement firing process optimizing method based on clinker quality index | |
CN103729569A (en) | Soft measurement system for flue gas of power-station boiler on basis of LSSVM (Least Squares Support Vector Machine) and online updating | |
CN106594794B (en) | A kind of hybrid intelligent boiler efficiency burning optimization model update method | |
CN110986085B (en) | Four-corner tangential boiler combustion optimization control method based on air distribution mode optimization | |
CN112765902A (en) | RBF neural network soft measurement modeling method based on TentFWA-GD and application thereof | |
CN114334025A (en) | Construction and verification method of cement clinker calcination environment variable | |
Xu et al. | A new online optimization method for boiler combustion system based on the data-driven technique and the case-based reasoning principle | |
Chen et al. | A novel machine learning-based model predictive control framework for improving the energy efficiency of air-conditioning systems | |
Xu et al. | On-line combustion optimization framework for coal-fired boiler combining improved cultural algorithm, deep learning, multi-objective evolutionary algorithm with improved case-based reasoning technology | |
Dong et al. | Evaluation of TPGU using entropy-improved TOPSIS-GRA method in China | |
Hu et al. | Dynamic modeling framework based on automatic identification of operating conditions for sintering carbon consumption prediction | |
CN113049259A (en) | Fuzzy control method of rack control system, storage medium and equipment |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190709 |