CN105279573B - A kind of thermal power plant's coa consumption rate Economic Analysis Method - Google Patents
A kind of thermal power plant's coa consumption rate Economic Analysis Method Download PDFInfo
- Publication number
- CN105279573B CN105279573B CN201510599277.5A CN201510599277A CN105279573B CN 105279573 B CN105279573 B CN 105279573B CN 201510599277 A CN201510599277 A CN 201510599277A CN 105279573 B CN105279573 B CN 105279573B
- Authority
- CN
- China
- Prior art keywords
- formula
- consumption rate
- model
- error
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of thermal power plant's coa consumption rate Economic Analysis Method, its main feature is that, it comprises the step of：The process variables such as main steam flow, coal-water ratio, reheater steam pressure, total blast volume, condenser vacuum, dry slag temperature, boiler furnace negative pressure, feedwater flow, reheater temperature and flue gas oxygen content are measured by related thermal measurement component；The real-time generated output of fired power generating unit is read by the power meter at scene, enters the total coal amount of stove divided by the real-time generated output of fired power generating unit by what is measured, obtains monoblock real-time coal consumption rate.It is input with above-mentioned process variable, is output with monoblock real-time coal consumption rate, principal component is extracted based on offset minimum binary theory；Then it is that coa consumption rate economic analysis model is built in output with coa consumption rate using the principal component extracted as mode input.The economy quality of test sample is judged according to power industry coa consumption rate evaluation criterion.With model prediction accuracy higher, prediction deviation meets engine request, and practical engineering application is worth the advantages that high.
Description
Technical field
The present invention relates to highly energy-consuming trade economic analysis field more particularly to points of coal unit coa consumption rate economy
Analysis, is a kind of thermal power plant's coa consumption rate Economic Analysis Method.
Background technology
Currently, in the electricity market in China, there is easy situation.Then, there is major electricity power group
The situation surfed the Net at a competitive price mutually.Since in this way, a higher requirement just has been proposed to the economy of power plant's operation.Then, respectively
The energy-saving of a thermal power plant has been mentioned a unprecedented height.And the economy of a thermal power plant is evaluated, directly
The embodiment connect is exactly coa consumption rate.
So-called coa consumption rate refers to thermal power plant and often produces or supply the Coal-fired capacity consumed needed for 1 kilowatt-hour of electric energy.It is logical
It commonly uses and consumes how many grams of standard coals per kilowatt hour to indicate.If Intelligent Information Processing knowwhy can be utilized, in time to thermoelectricity
Factory's coa consumption rate is made prediction, and is made unit operation adjustment in advance, can not only be effectively reduced power plant's coal consumption, cost-effective, and
Positive meaning is also played to the protection of non-renewable resources and environment.With a rated load for 350MW supercritical thermal powers
It for unit, if often degree electric energy enough reduces 1 gram of standard coal of consumption, is run 300 days according to annual single unit, average unit is negative
Lotus 300MW is calculated, and over a year, can save 2160 tons of standard coal, reduces nuisance (mainly dust, CO2And SO2) discharge
38.3 tons；It is calculated according to 700 yuan of mark coal per ton, cost-saved more than 150 ten thousand yuan.This is only the accounting result of a unit.According to
Official statistics, China's whole year gross generation in 2014 are 5,463,800,000,000 kilowatt hours, and thermoelectricity class is by 70% conversion, then 2014 national
Degree fired power generating unit generated energy is 38246.6 hundred million kilowatt hours, and under the premise of every degree electricity still saves 1 gram of standard coal, whole year can be saved
About 3,824,660 tons of coal of mark, cost-saved about 26.77 hundred million yuan.If these coals are transported with train, by often section compartment capacity
It it is 60 tons, each column train 40 saves compartment and calculates, and needs about 1600 row trains in total, it is clear that railway can be effectively relieved in this behave
The pressure of traffic.And to the influence of environment, then it can not be weighed with money.It can be seen that coa consumption rate is to society, economy, ring
The influence in border etc. has some idea of.Due to existing coa consumption rate analysis model, accuracy rate is not generally high.Here, quasi- by surveying in real time
Data unit operation is measured, partial least squares algorithm and nonlinear regression vector machine are based on, proposes that a kind of judge accuracy rate is higher
Thermal power plant's coa consumption rate Economic Analysis Method.
Invention content
Technical problem solved by the invention is：A kind of scientific and reasonable, strong applicability is proposed, convenient for grasping, analysis is accurate
True thermal power plant's coa consumption rate Economic Analysis Method based on partial least squares algorithm and nonlinear regression vector machine, this method is only
Real-time measuring unit operation data is needed, trained analysis model is substituted into, unit coa consumption rate situation can be directly predicted, be
Economy of power plant analysis provides foundation.
Solving the scheme of its technical problem is：A kind of thermal power plant's coa consumption rate Economic Analysis Method, characterized in that it includes
Following steps：
1) acquisition of data：
Main steam flow, coal-water ratio, reheater steam pressure, total blast volume, condensing are measured by related thermal measurement component
10 device vacuum, dry slag temperature, boiler furnace negative pressure, feedwater flow, reheater temperature and flue gas oxygen content process variables, specifically
It measures as follows：
Main steam flow：It is measured using HD-W insert type vortex flow meters, t/h；
Fuel- Water Rate：Boiler feed capacity is measured by AKS-LDC intelligence inserted electromagnet flow meters respectively, matches the F55 pressure resistances that rub
Gravimetric Coal Feeders measure furnace coal weight, are then obtained by calculating the ratio of the former with the latter, scalar；
Reheater steam pressure：It is measured using PPM-T322B pressure transmitters, MPa；
Total blast volume：Exempt from purge air quantity measuring device using FS-5C block-resistant types to measure, t/h；
Condenser vacuum：It is measured using PDM-520 Miniature precision resistance vacuum gauges, KPa；
Dry slag temperature：It is measured using infrared radiation thermometer IS-CF1400AD, DEG C；
Combustion chamber draft：It is measured using BTS6800-SP negative pressure pick-up devices, Pa；
Feedwater flow：It is measured by AKS-LDC intelligence inserted electromagnet flow meters, t/h；
Reheater steam temperature：It is measured using WSSXP-401 bimetallic thermometers, DEG C；
Flue gas oxygen content：It is measured using XP-3180 oxygen content testing instruments, %；
2) extraction of principal component
1. data normalization
In order to extract principal component, first by independent variable matrix
X=(xij)n×m (1)
In formula, n indicates selected training sample dosage, group；
M indicates selected Heat-work parameter dimension, in this m=1,2 ..., 10；
I=1,2 ..., n；
J=1,2 ..., m
With dependent variable matrix
Y=(yij)n×p (2)
In formula, n indicates selected training sample dosage；
P indicates dependent variable dimension, in this p=1；
I=1,2 ..., n；
J=1,2 ..., p
It is standardized using following formula (3), (4), the set center of gravity of sample point is made to be overlapped with coordinate origin：
In formula, E1、F1The normalized matrix of respectively X and Y；
E(xi), E (y) be respectively X, Y mean value；
SyThe respectively mean square deviation of X, Y；
2. first constituents extraction t1：
The combination coefficient w of the 1st step is calculated using following formula (5)1：
First principal component t is extracted using following formula (6)1：
t1=E1w1 (6)
3. calculating residual matrix
Residual matrix is calculated using following formula (7), (8)：
E2=E1-t1P1 T (7)
F2=F1-t1r1 (8)
In formula,For regression coefficient, vector；
r1=F1 Tt1/||t1||2, it is regression coefficient, scalar；
4. CALCULATING PREDICTION residual sum of squares (RSS) PRESS
Prediction error is calculated using following formula (9)：
In formula, yiFor coa consumption rate actual value, g/kwh；
y1(-i)It is in extraction principal component t1Under the premise of, utilize the remaining sample for removing i-th of sample (i=1,2....n)
This, seeks regression equation, i-th of sample point is then substituted into the obtained prediction equation value of the regression equation, g/kwh；
5. utilizing residual matrix E2、F2Instead of step 2. in E1、F13. 4., repeat step 2., continue to extract principal component；
6. in being drawn 5. using rectangular coordinate system " obtained predicted residual quadratic sum value is recycled every time --- cycle time
Number " curve graph determines the cycle-index so that corresponding when predicted residual quadratic sum PRESS is minimized by the coordinate diagram,
Namely best Principle component extraction number q；
7. according to best Principle component extraction number, q principal component is extracted, constitutes the matrix A of q × n dimensions；
3) nonlinear regression vector machine model is built
With matrix Aq×nFor input variable, it is output with coa consumption rate, builds nonlinear regression vector machine model, detailed process
It is as follows：
(a) selection of kernel function
Here, selecting kernel function of the Radial basis kernel function as vector machine, Radial basis kernel function expression formula is：
(b) training error formula
Using sample model is trained during, for ensure model the training end time and training error most
Small, using average relative error, calculation formula is：
In formula, n is training sample dosage, group；
yiFor model predication value, g/kwh；
yi0For practical coa consumption rate, g/kwh；
(c) steps are as follows for model buildings：
Model prediction exports expression formula：
In formula, ψ (x, xi) it is support vector machines kernel function, it is herein Radial basis kernel function；
In formula, α is by the α in formula (12)iThe matrix-vector of composition；
B=Ω+γ-1I (15)
In formula, I is unit vector；γ is penalty coefficient；Matrix-vector Ω is calculated using following formula (16)：
Ωkj=Ψ (xk,xj) (k, j=1,2 ..., n) (16)
In formula, ψ (xk,xj) it is support vector machines kernel function, it is herein Radial basis kernel function；N is training sample dosage,
Formula (13)-(16) are substituted into formula (12) and obtain final mask prediction output；
4) training of model
In order to improve model training speed, rapidly and accurately determines penalty coefficient, the core coefficient optimum combination of model, reach
The target training error of model, here, during carrying out optimizing to penalty coefficient and core coefficient, using step in detail below：
(1) a point M is arbitrarily determined in rectangular coordinate system first quartile1(a1,b1), a is taken under normal conditions1、b1>0；
(2) with M1Centered on point, with 2c1For short diagonal, 2d1For long-diagonal, a diamond shape T is drawn1, diamond shape T1Pair
Linea angulata is respectively perpendicular to the horizontal x of reference axis, y-axis or y, x-axis, wherein c1<min(a1,b1), d1<max(a1,b1)；
(3) with diamond shape T1Four vertex together with point M coordinate as five of penalty coefficient-core coefficient combinations, respectively
Model is substituted into be trained；
(4) the error amount e corresponding to each point is calculated by training error formula1,e2,…,ei, enable e=min { e1,e2,…,
ei(i=1,2 ..., 5)；
(5) when assuming that error minimum value is e, the point of corresponding point is M2(a2,b2), and point M2Place diamond shape T1Diagonal line
Length be l, then with point M2Centered on, according to the method for step 2., draw another diamond shape T2, wherein a2<L/2, b2<l/2；
(6) with diamond shape T2Four apex coordinates be four of penalty coefficient-core coefficient combinations, substitute into model respectively and carry out
Training, calculates respective training error, and and M2The error of point is compared；
(7) if the error of four points is all more than M2The error of point then changes the catercorner length of diamond shape, repaints diamond shape,
Repeat above (2)-(6) step；If having less than M in the error of four points2Point, then above step is repeated centered on new point
(5) and (6) and so on preset error range until meeting；
5) coa consumption rate prediction and economic analysis
Sample to be tested is inputted trained model to predict, and according to the requirement according to power industry, it will be pre-
It is standard coal coa consumption rate to survey result conversion, according to power industry coa consumption rate evaluation criterion to the economy quality of the test sample into
Row is judged.
The present invention is based on the coa consumption rate economic analysis sides of thermal power plant of partial least squares algorithm and nonlinear regression vector machine
Method is compared with numerous domestic and international coa consumption rate analysis models, and the coupling model precision of prediction higher, prediction deviation, which meets engineering, to be wanted
It asks, therefore institute's established model has practical engineering application value.With scientific and reasonable, strong applicability, convenient for grasping, analyze accurate etc. excellent
Point.
Description of the drawings
The data collecting system figure that Fig. 1 is built for the present invention；
In figure：1HD-W insert type flux of vortex street instrument, 2AKS-LDC intelligence inserted electromagnet flow meters, 3 match the F55 pressure resistances that rub
Gravimetric Coal Feeders coal measuring point, 4PPM-T322B pressure transmitters, 5FS-5C block-resistant types exempt from purge air quantity measuring device, 6PDM-
520 Miniature precision resistance vacuum gauges, 7 infrared radiation thermometer IS-CF1400AD, 8BTS6800-SP negative pressure pick-up devices, 9WSSXP-
401 bimetallic thermometers, 10XP-3180 oxygen content testing instruments, 11 grind magnificent PCI1710 data collecting cards, 12IBM X3800 services
Device, 13 thermal power plant's coa consumption rate economic analysis systems.
Fig. 2 is the model of the invention built to the prediction result figure of partial test sample；
Fig. 3 is the model of the invention built to as-fired coal kind mixed-fuel burning proportion Economic Evaluation figure.
Specific implementation mode
Present invention will be further explained below with reference to the attached drawings and examples.
A kind of thermal power plant's coa consumption rate Economic Analysis Method of the present invention, it is comprised the following steps：
1) start thermal power plant's data collecting system, and the main steaming of fired power generating unit is measured by HD-W insert type flux of vortex street instrument 1
Stripping temperature；Measure boiler feed capacity by AKS-LDC intelligence inserted electromagnet flow meter 2, and by match rub F55 pressure resistances weigh to
Coal machine measuring point 3 measures furnace coal weight, and the coal amount of all feeders of the monoblock is summed the total coal amount of stove, uses boiler later
Feedwater flow divided by enter the total coal amount of stove, obtains ratio of water to coal (being commonly called as coal-water ratio)；It is measured again using PPM-T322B pressure transmitters 4
Hot steam pressure；Exempt from purge air quantity measuring device 5 using FS-5C block-resistant types to measure into stove total blast volume；Using the miniature essences of PDM-520
Close resistance vacuum gauge 6 measures condenser vacuum；The dry slag temperature of boiler is measured using IS-CF1400AD infrared radiation thermometers 7；It adopts
Boiler furnace negative pressure is measured with BTS6800-SP negative pressure pick-up device 8；Reheating is measured using WSSXP-401 bimetallic thermometers 9
The temperature of device steam；Boiler tail flue gas oxygen content is measured using XP-3180 oxygen content testing instruments 10.Pass through the power meter at scene
The real-time generated output of the fired power generating unit is read, will be removed by matching the total coal amount of stove that enters that F55 pressure resistance Gravimetric Coal Feeders measuring point 3 measures of rubbing
With the real-time generated output of the fired power generating unit, you can obtain the monoblock real-time coal consumption rate.The monoblock real-time coal consumption rate is connected
With, by grinding collected 10 parameters of the institute of magnificent PCI1710 data collecting cards 11, being all delivered to IBM X3800 services in step 1
Device 12 starts thermal power plant's coa consumption rate economic analysis system 13.Later, it is input with 10 parameters, is with real-time coal consumption rate
Output, extracts principal component, and detailed process is as follows：
2) extraction of principal component
1. data normalization
In order to extract principal component, first by independent variable matrix
X=(xij)n×m (1)
In formula, n indicates selected training sample dosage, group；
M indicates selected Heat-work parameter dimension, in this m=1,2 ..., 10；
I=1,2 ..., n；
J=1,2 ..., m
With dependent variable matrix
Y=(yij)n×p (2)
In formula, n indicates selected training sample dosage；
P indicates dependent variable dimension, in this p=1；
I=1,2 ..., n；
J=1,2 ..., p
It is standardized using following formula (3), (4), the set center of gravity of sample point is made to be overlapped with coordinate origin：
In formula, E1、F1The normalized matrix of respectively X and Y；
E(xi), E (y) be respectively X, Y mean value；
SyThe respectively mean square deviation of X, Y；
2. first constituents extraction t1：
The combination coefficient w of the 1st step is calculated using following formula (5)1：
First principal component t is extracted using following formula (6)1：
t1=E1w1 (6)
3. calculating residual matrix
Residual matrix is calculated using following formula (7), (8)：
E2=E1-t1P1 T (7)
F2=F1-t1r1 (8)
In formula,For regression coefficient, vector；
r1=F1 Tt1/||t1||2, it is regression coefficient, scalar；
4. CALCULATING PREDICTION residual sum of squares (RSS) PRESS
Prediction error is calculated using following formula (9)：
In formula, yiFor coa consumption rate actual value, g/kwh；
y1(-i)It is in extraction principal component t1Under the premise of, utilize the remaining sample for removing i-th of sample (i=1,2....n)
This, seeks regression equation, i-th of sample point is then substituted into the obtained prediction equation value of the regression equation, g/kwh；
5. utilizing residual matrix E2、F2Instead of step 2. in E1、F13. 4., repeat step 2., continue to extract principal component；
6. in being drawn 5. using rectangular coordinate system " obtained predicted residual quadratic sum value is recycled every time --- cycle time
Number " curve graph determines the cycle-index so that corresponding when predicted residual quadratic sum PRESS is minimized by the coordinate diagram,
Namely best Principle component extraction number q；
7. according to best Principle component extraction number, q principal component is extracted, constitutes the matrix A of q × n dimensions；
3) nonlinear regression vector machine model is built
With matrix Aq×nFor input variable, it is output with coa consumption rate, builds nonlinear regression vector machine model, detailed process
It is as follows：
(a) selection of kernel function
Here, selecting kernel function of the Radial basis kernel function as vector machine, Radial basis kernel function expression formula is：
(b) training error formula
Using sample model is trained during, for ensure model the training end time and training error most
Small, using average relative error, calculation formula is：
In formula, n is training sample dosage, group；
yiFor model predication value, g/kwh；
yi0For practical coa consumption rate, g/kwh；
(c) steps are as follows for model buildings：
Model prediction exports expression formula：
In formula, ψ (x, xi) it is support vector machines kernel function, it is herein Radial basis kernel function；
In formula, α is by the α in formula (12)iThe matrix-vector of composition；
B=Ω+γ-1I (15)
In formula, I is unit vector；γ is penalty coefficient；Matrix-vector Ω is calculated using following formula (16)：
Ωkj=Ψ (xk,xj) (k, j=1,2 ..., n) (16)
In formula, ψ (xk,xj) it is support vector machines kernel function, it is herein Radial basis kernel function；N is training sample dosage,
Formula (13)-(16) are substituted into formula (12) and obtain final mask prediction output；
4) training of model
In order to improve model training speed, rapidly and accurately determines penalty coefficient, the core coefficient optimum combination of model, reach
The target training error of model, here, during carrying out optimizing to penalty coefficient and core coefficient, using step in detail below：
(1) a point M is arbitrarily determined in rectangular coordinate system first quartile1(a1,b1), a is taken under normal conditions1、b1>0；
(2) with M1Centered on point, with 2c1For short diagonal, 2d1For long-diagonal, a diamond shape T is drawn1, diamond shape T1Pair
Linea angulata is respectively perpendicular to the horizontal x of reference axis, y-axis or y, x-axis, wherein c1<min(a1,b1), d1<max(a1,b1)；
(3) with diamond shape T1Four vertex together with point M coordinate as five of penalty coefficient-core coefficient combinations, respectively
Model is substituted into be trained；
(4) the error amount e corresponding to each point is calculated by training error formula1,e2,…,ei, enable e=min { e1,e2,…,
ei(i=1,2 ..., 5)；
(5) when assuming that error minimum value is e, the point of corresponding point is M2(a2,b2), and point M2Place diamond shape T1Diagonal line
Length be l, then with point M2Centered on, according to the method for step 2., draw another diamond shape T2, wherein a2<L/2, b2<l/2；
(6) with diamond shape T2Four apex coordinates be four of penalty coefficient-core coefficient combinations, substitute into model respectively and carry out
Training, calculates respective training error, and and M2The error of point is compared；
(7) if the error of four points is all more than M2The error of point then changes the catercorner length of diamond shape, repaints diamond shape,
Repeat above (2)-(6) step；If having less than M in the error of four points2Point, then above step is repeated centered on new point
(5) and (6) and so on preset error range until meeting；
5) coa consumption rate prediction and economic analysis
Sample to be tested is inputted into trained model and predicts that model is to partial test sample predictions result such as Fig. 2
It is shown.Model worst error -0.61g/kwh, mean error (after taking absolute value) 0.18g/kwh.In instantly national hair
Under the overall background that electric every gram of coal of enterprise's coal consumption must be striven, it is clear that coupling model has preferably extensive compared with other two single models
Ability, accuracy meet engineering requirements, are favored by all power plant.
It is standard coal coa consumption rate by prediction result conversion according to the requirement according to power industry.According to power industry coal consumption
Rate evaluation criterion judges the economy quality of the test sample.To economy evaluation result such as Fig. 3 institutes of part sample
Show.As seen from the figure, in order to which the economy obtained, correction coal 1 and the mixed-fuel burning proportion for correcting coal 2 should be not less than 5:1, it is best
It can control 6:1 or more.
Claims (1)
1. a kind of thermal power plant's coa consumption rate Economic Analysis Method, characterized in that it is comprised the following steps：
1) acquisition of data
Main steam flow, coal-water ratio, reheater steam pressure, total blast volume, condenser are measured by related thermal measurement component
10 vacuum, dry slag temperature, boiler furnace negative pressure, feedwater flow, reheater steam temperature and flue gas oxygen content process variables, tool
Bulk measurement is as follows：
Main steam flow：It is measured using HD-W insert type vortex flow meters, t/h；
Coal-water ratio：Boiler feed capacity is measured by AKS-LDC intelligence inserted electromagnet flow meters respectively, the F55 pressure resistances that rub is matched and weighs
Feeder measures furnace coal weight, is then obtained by calculating the ratio of the former with the latter, scalar；
Reheater steam pressure：It is measured using PPM-T322B pressure transmitters, MPa；
Total blast volume：Exempt from purge air quantity measuring device using FS-5C block-resistant types to measure, t/h；
Condenser vacuum：It is measured using PDM-520 Miniature precision resistance vacuum gauges, KPa；
Dry slag temperature：It is measured using infrared radiation thermometer IS-CF1400AD, DEG C；
Combustion chamber draft：It is measured using BTS6800-SP negative pressure pick-up devices, Pa；
Feedwater flow：It is measured by AKS-LDC intelligence inserted electromagnet flow meters, t/h；
Reheater steam temperature：It is measured using WSSXP-401 bimetallic thermometers, DEG C；
Flue gas oxygen content：It is measured using XP-3180 oxygen content testing instruments, %；
2) extraction of principal component
1. data normalization
In order to extract principal component, first by independent variable matrix
X=(xij)n×m (1)
In formula, n indicates selected training sample dosage, group；
M indicates selected Heat-work parameter dimension, in this m=1,2 ..., 10；
X indicates that training sample concentrates each process variable numerical value of independent variable；
I=1,2 ..., n；
J=1,2 ..., m
With dependent variable matrix
Y=(yij)n×p (2)
In formula, n indicates selected training sample dosage；
P indicates dependent variable dimension；
Y indicates that training sample concentrates the coa consumption rate actual value corresponding to each group of independent variable, g/kwh；
I=1,2 ..., n；
J=1,2 ..., p, here, p=1,
It is standardized using following formula (3), (4), the set center of gravity of sample point is made to be overlapped with coordinate origin：
In formula, E1、F1The normalized matrix of respectively X and Y；
E(xi), E (y) be respectively X, Y mean value；
SyThe respectively mean square deviation of X, Y；
2. first constituents extraction t1：
The combination coefficient w of the 1st step is calculated using following formula (5)1：
First principal component t is extracted using following formula (6)1：
t1=E1w1 (6)
3. calculating residual matrix
Residual matrix is calculated using following formula (7), (8)：
E2=E1-t1P1 T (7)
F2=F1-t1r1 (8)
In formula,For regression coefficient, vector；
r1=F1 Tt1/||t1||2, it is regression coefficient, scalar；
4. CALCULATING PREDICTION residual sum of squares (RSS) PRESS
Prediction error is calculated using following formula (9)：
In formula, yiFor coa consumption rate actual value, g/kwh；
y1(-i)It is in extraction principal component t1Under the premise of, using removing i-th of sample, i=1,2 ... .n, remaining sample, ask
Then i-th of sample point is substituted into the obtained prediction equation value of the regression equation, g/kwh by regression equation；
5. utilizing residual matrix E2、F2Instead of step 2. in E1、F13. 4., repeat step 2., continue to extract principal component；
6. using rectangular coordinate system draw 2. -5. " recycle obtained predicted residual quadratic sum PRESS every time in cyclic process
Value --- cycle-index " curve graph is determined by the coordinate diagram so that institute is right when predicted residual quadratic sum PRESS is minimized
The cycle-index answered, namely best Principle component extraction number q；
7. according to best Principle component extraction number, q principal component is extracted, constitutes the matrix A of q × n dimensions；
3) nonlinear regression vector machine model is built
With matrix Aq×nFor input variable, it is output with coa consumption rate, builds nonlinear regression vector machine model, detailed process is as follows：
(a) selection of kernel function
Here, selecting kernel function of the Radial basis kernel function as vector machine, Radial basis kernel function expression formula is：
In formula,σ is core coefficient；
(b) training error formula
During being trained to model using sample, to ensure training end time and the training error minimum of model, adopt
With average relative error, calculation formula is：
In formula, n is training sample dosage, group；
yiFor model predication value, g/kwh；
yi0For practical coa consumption rate, g/kwh；
(c) steps are as follows for model buildings：
Model prediction exports expression formula：
In formula, ψ (x, xi) it is support vector machines kernel function, it is herein Radial basis kernel function；
In formula, α is by the α in formula (12)iThe matrix-vector of composition；
In formula,
B=Ω+γ-1I (15)
In formula, I is unit vector；γ is penalty coefficient；Matrix-vector Ω is calculated using following formula (16)：
Ωkj=Ψ (xk,xj), k, j=1,2 ..., n (16)
In formula, ψ (xk,xj) it is support vector machines kernel function, it is herein Radial basis kernel function；N is training sample dosage,
Formula (13)-(16) are substituted into formula (12) and obtain final mask prediction output；
4) training of model
In order to improve model training speed, rapidly and accurately determines penalty coefficient, the core coefficient optimum combination of model, reach model
Target training error, here, during carrying out optimizing to penalty coefficient and core coefficient, using step in detail below：
(1) a point M is arbitrarily determined in rectangular coordinate system first quartile1(a1,b1), a is taken under normal conditions1、b1>0；
(2) with M1Centered on point, with 2c1For short diagonal, 2d1For long-diagonal, a diamond shape T is drawn1, diamond shape T1Diagonal line
It is respectively perpendicular to the horizontal x of reference axis, y-axis or y, x-axis, wherein c1＜ min (a1,b1), d1＜ max (a1,b1)；
(3) with diamond shape T1Four vertex together with point M1Coordinate as five of penalty coefficient-core coefficient combinations, substitute into respectively
Model is trained；
(4) the error amount e corresponding to each point is calculated by training error formula1,e2,…,ei, enable e=min { e1,e2,…,ei,
I=1,2 ..., 5；
(5) when assuming that error minimum value is e, the point of corresponding point is M2(a2,b2), and point M2Place diamond shape T1Cornerwise length
Degree is l, then with point M2Centered on, according to the method for step (2), draw another diamond shape T2, wherein a2＜ l/2, b2＜ l/2；
(6) with diamond shape T2Four apex coordinates be four of penalty coefficient-core coefficient combinations, substitute into model respectively and be trained,
Calculate respective training error, and and M2The error of point is compared；
(7) if the error of four points is all more than M2The error of point then changes the catercorner length of diamond shape, repaints diamond shape, repeats
(2)-(6) step above；If having less than M in the error of four points2Point, then above step (5) is repeated centered on new point
(6), and so on, error range is preset until meeting；
5) coa consumption rate prediction and economic analysis
Sample to be tested is inputted trained model to predict, and according to the requirement according to power industry, prediction is tied
Fruit conversion is standard coal coa consumption rate, is commented the economy quality of the test sample according to power industry coa consumption rate evaluation criterion
Sentence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510599277.5A CN105279573B (en) | 2015-09-19 | 2015-09-19 | A kind of thermal power plant's coa consumption rate Economic Analysis Method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510599277.5A CN105279573B (en) | 2015-09-19 | 2015-09-19 | A kind of thermal power plant's coa consumption rate Economic Analysis Method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105279573A CN105279573A (en) | 2016-01-27 |
| CN105279573B true CN105279573B (en) | 2018-09-07 |
Family
ID=55148549
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510599277.5A Active CN105279573B (en) | 2015-09-19 | 2015-09-19 | A kind of thermal power plant's coa consumption rate Economic Analysis Method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105279573B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679018A (en) * | 2017-08-16 | 2018-02-09 | 浙江浙能富兴燃料有限公司 | A kind of poor computational methods of coal-burning power plant limit coal unit price |
| CN112258019A (en) * | 2020-10-19 | 2021-01-22 | 佛山众陶联供应链服务有限公司 | Coal consumption assessment method |
| CN112510703B (en) * | 2020-11-26 | 2022-10-04 | 贵州电网有限责任公司 | Multi-energy access power grid optimal scheduling method considering coal consumption curve correction |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103440528A (en) * | 2013-08-12 | 2013-12-11 | 国家电网公司 | Thermal power generating unit operation optimization method and device based on consumption difference analysis |
| CN103995467A (en) * | 2014-05-26 | 2014-08-20 | 河海大学常州校区 | Method for extracting main components of dredging operation energy consumption influence factors based on partial least squares |
-
2015
- 2015-09-19 CN CN201510599277.5A patent/CN105279573B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103440528A (en) * | 2013-08-12 | 2013-12-11 | 国家电网公司 | Thermal power generating unit operation optimization method and device based on consumption difference analysis |
| CN103995467A (en) * | 2014-05-26 | 2014-08-20 | 河海大学常州校区 | Method for extracting main components of dredging operation energy consumption influence factors based on partial least squares |
Non-Patent Citations (2)
| Title |
|---|
| Research on Forecast Model for Sustainable Development of Economy-Environment System Based on PCA and SVM;Yan Li etc.;《Proceedings of the Fifth International Conference on Machine Learning and Cybernetics》;20090304;第3590-3593页 * |
| 基于偏最小二乘回归分析的供电煤耗研究;刘建华 等;《电力科学与工程》;20110428;第27卷(第4期);第30-33页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105279573A (en) | 2016-01-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Han et al. | Thermodynamic analysis and life cycle assessment of supercritical pulverized coal-fired power plant integrated with No. 0 feedwater pre-heater under partial loads | |
| US20190113417A1 (en) | Method for acquiring thermal efficiency of a boiler | |
| CN105224735B (en) | Generating set energy efficiency analysis method for air | |
| CN109785187B (en) | Method for correcting power supply coal consumption detection data of generator set | |
| CN105279573B (en) | A kind of thermal power plant's coa consumption rate Economic Analysis Method | |
| CN101697179A (en) | Method for measuring and calculating trend of heat value of fuel coal of power station boiler based on positive and negative heat balance relationship | |
| CN102803847A (en) | Method for determination of carbon dioxide emissions from steam generation systems | |
| CN106018730B (en) | Ature of coal device for measuring moisture and method based on coal pulverizer inlet First air amendment | |
| CN112131517B (en) | Method for measuring and calculating lower calorific value of garbage in garbage incineration power plant | |
| CN108197723B (en) | Optimized energy-saving scheduling method for coal consumption and pollutant discharge of coal-electricity unit power supply | |
| CN104732451A (en) | Low-pressure economizer energy saving assessment method applied to power plant thermal system | |
| US7398652B1 (en) | System for optimizing a combustion heating process | |
| CN106960113A (en) | A kind of divisions of responsibility method on single shaft combined cycle generating unit performances acceptance Zhong Ji Dao Yulu islands | |
| CN103816987A (en) | Method for calculating powder output of double-inlet and double-outlet coal mill | |
| CN108182553A (en) | A kind of coal-fired boiler combustion efficiency On-line Measuring Method | |
| CN108595723A (en) | A kind of Boiler Air Heater's time heat Calculation method and device | |
| CN109934493B (en) | Method for rapidly determining coal consumption characteristic curve of thermal generator set | |
| CN103728055B (en) | A kind of real-time estimation method of thermal power unit boiler furnace outlet flue gas energy | |
| CN103697958B (en) | The real time measure method of coal unit drum outlet saturation steam mass rate | |
| CN109442465B (en) | Fitting sample selection and metering method for instantaneous powder feeding amount of pulverized coal boiler | |
| CN103699790B (en) | A kind of real-time detection method of coal fired power plant furnace outlet flue gas mean temperature | |
| CN108875165B (en) | Boiler characteristic calibration method based on operation data | |
| CN102944441A (en) | Method for outputting effective output efficiency of boiler | |
| CN103699782B (en) | Coal feeding amount soft measuring method of middle-speed powder milling and preparation system | |
| CN207112772U (en) | Boiler hearth temperature on-line prediction system based on core offset minimum binary |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |