CN106650191B - Wind farm power prediction screening sample method based on dual confidence interval - Google Patents
Wind farm power prediction screening sample method based on dual confidence interval Download PDFInfo
- Publication number
- CN106650191B CN106650191B CN201510729241.4A CN201510729241A CN106650191B CN 106650191 B CN106650191 B CN 106650191B CN 201510729241 A CN201510729241 A CN 201510729241A CN 106650191 B CN106650191 B CN 106650191B
- Authority
- CN
- China
- Prior art keywords
- confidence interval
- sample
- screening
- weight
- days
- 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
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Landscapes
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Wind farm power prediction screening sample method disclosed by the invention based on dual confidence interval, including step：The bound that two linear functions are arranged first as the first weight confidence interval rejects all samples except this confidence interval according to the sieveing coeffecient of setting；Secondly according to the active relationship of wind speed-of wind power plant, bound of two nonlinear piecewise functions as the second weight confidence interval is set, reject all samples except this confidence interval, the bound function coefficients of the second weight confidence interval are constantly adjusted simultaneously, ensure to fall after the number of samples of confidence interval accounts for sieveing coeffecient of the total sample number purpose ratio value more than or equal to setting, completes screening sample.The present invention, which can filter out, can most reflect the active sample of the wind speed-of wind field operation characteristic, correct caused by conventional method the shortcomings that training does not restrain and be easily introduced the wind speed-active mapping relations of mistake.
Description
Technical field
The present invention relates to wind farm power prediction fields, specifically based on dual confidence interval in wind farm power prediction
Wind farm power prediction screening sample method.
Background technology
The construction of wind power plant is more and more at present, and wind farm power prediction is link important in wind power plant.
Current wind farm power prediction generally using intelligent algorithm, input wind power plant actual measurement air speed data and
Active data are surveyed, the active mapping relations of wind speed-of wind power plant are obtained after being trained, then carry out power prediction.Using artificial
Intelligent algorithm is trained the active data of wind speed-, has very strong dependence to wind speed, this active two classes power prediction sample.
Traditional method is obtained by all actual measurement wind speed for acquiring wind power plant and all active data of actual measurement after rejecting data quality exception
It is trained, has the disadvantage that the data filtered out, then using intelligent algorithm：
Training does not restrain：The small wind in part and ration the power supply maintenance down when the active sample of wind speed-can not reject so that filter out
Sample be in divergent state, can not be restrained using neural metwork training, the active mapping relations of wind speed-can not be obtained, caused
It can not carry out wind farm power prediction.
Introduce wrong mapping relations：It is convergent in neural metwork training after part abnormal data introduces, it introduces
The wind speed of mistake-power mapping relations, leads to wind farm power prediction results abnormity.
Invention content
In order to overcome traditional wind farm power prediction screening sample method to lead to the trained wind speed-that do not restrain, introduce mistake
A kind of defect of active mapping relations, it is proposed that wind power prediction screening sample method based on dual confidence interval.
Two linear functions are arranged as the first heavy confidence interval in the screening sample of wind farm power prediction of the present invention first
Bound all samples except this confidence interval are rejected according to the sieveing coeffecient of setting；Secondly according to the wind of wind power plant
Speed-active relationship, bound of two nonlinear piecewise functions of setting as the second weight confidence interval, is rejected in this confidence interval
Except all samples, while constantly adjustment second weight confidence interval bound function coefficients, guarantee fall in confidence interval
After number of samples accounts for sieveing coeffecient of the total sample number purpose ratio value more than or equal to setting, screening sample is completed.
The purpose of the present invention is reached by following measure：Wind farm power prediction sample based on dual confidence interval
Screening technique includes the following steps：
Step (1), initialization sample select number of days；
Step (2), it is active to reality to be normalized, establishes two-dimensional coordinate so that the reality after wind speed and normalization is active
System；
Step (3), the bound function of initialization the first weight confidence interval, the first heavy confidence interval as screening sample
Bound, it is ensured that fall first weight confidence interval in sample number >=total sample number * sieveing coeffecients a1, increase if being unsatisfactory for
It is loaded this number of days, to obtain enough sample numbers, it is ensured that this part sample is scolded in the first resetting letter area as screening sample
In；
Step (4), carries out the screening sample of the second weight confidence interval, and the second heavy confidence interval bound function of setting carries out
Screening is screened if number of samples is unsatisfactory for expected requirement after adjusting bound function again；If at this time beyond adjustment
Expected requirement is still cannot be satisfied, sample number of days is increased, until sample number meets expected require；
Step (5), extracts the sample filtered out, and active P is multiplied by by normalizedInstallationRenormalization is carried out, is completed
Screening.
Further, in the step (1), initialization sample number of days selects 15 days or one week integral multiples.
Further, initialization sample number of days was 2 to 12 weeks.Initialization sample number of days is rule of thumb typically chosen 15 days,
Usually i.e. one week 7 days integral multiple can need to be set as differing in 2 to 8 weeks according to scene, typically not greater than 3 months.
Further, in the step (3)：The upper limit function of initialization the first weight confidence interval:ymax=kmax* x is under
Limit function:ymin=kmin* x, the bound of the first weight confidence interval as screening sample, wherein ymin=kmin* x is the first weight
Lower limit of confidence interval function；ymax=kmax* x is the first heavy confidence interval upper limit function；If fallen in the first weight confidence interval
Sample number<Total sample number * sieveing coeffecients a1, increase sample number of days, carry out sample collection after increasing number of days 2 every time, then carry out
Screening；If falling sample number >=total sample number * sieveing coeffecients a in the first weight confidence interval1, then first again screening finish；
Sieveing coeffecient a1Generally between 0.6-0.9.
Further, in the step (4)：When carrying out the second weight confidence interval screening, the second heavy confidence interval is initialized
Upper limit function：fmaxAnd lower limit function：fmin, the upper and lower of confidence interval is weighed using the two functions as the second of screening sample
Limit carries out next step screening sample；fmaxFor the second heavy confidence interval upper limit function, fminFor the second heavy lower limit of confidence interval letter
Number, later initialization dynamically adjust number t=0；If dynamic adjustment number t>tmax(dynamic adjustment number maximum value), increases
Sample number of days carries out screening sample after increasing number of days 2 every time；If dynamic adjustment number t≤tmax, tmaxDynamically to adjust number
Maximum value, judges whether the sample number fallen in the second weight confidence interval is more than or equal to total sample number * sieveing coeffecients a2If
It is that the then screening of the second weight confidence interval is completed；If it is not, dynamic the second heavy confidence interval upper limit function coefficient k of adjustmentmax
=kmax+△kmax, kmaxFor upper limit function coefficient, △ kmaxStep-length, adjustment are corrected for the second heavy confidence interval upper limit function coefficient
Upper limit function parameter cmax=cmax+△cmax, cmaxFor upper limit function parameter, △ cmaxFor the second heavy confidence interval upper limit function ginseng
Number corrects step-length；Adjust the second heavy lower limit of confidence interval function coefficients kmin=kmin-△kmin, kminFor the coefficient of lower limit function,
△kminStep-length, adjustment lower limit function parameter c are corrected for the second heavy lower limit of confidence interval function coefficientsmin=cmin-△cmin, cmin
The parameter of lower limit function, △ cminStep-length is corrected for the second heavy lower limit of confidence interval function parameter, after often adjustment is primary, dynamic is adjusted
Whole number t=t+1, then the second heavy screening sample is carried out, until falling sample number >=total sample number * in the second weight confidence interval
Sieveing coeffecient a2, then the screening sample in the second weight confidence interval is completed.
Advantageous effect
By obtaining the correct active mapping relations of wind speed-, prediction model is established, it is accurate to improve the active prediction of wind power plant
On the one hand rate is conducive to power grid reasonable arrangement the whole network planned production, is on the other hand conducive to wind power plant enterprise in grid company
Good economic benefit is obtained in performance appraisal.
Description of the drawings
Fig. 1 is the wind farm power prediction screening sample method logic diagram based on dual confidence interval.
Specific implementation mode
Include the following steps based on the wind farm power prediction screening sample method of dual confidence interval referring to Fig. 1：
Step (1), initialization sample select number of days；
Step (2), it is active to reality to be normalized, establishes two-dimensional coordinate so that the reality after wind speed and normalization is active
System；
Step (3), the bound function of initialization the first weight confidence interval, the first heavy confidence interval as screening sample
Bound, it is ensured that fall first weight confidence interval in sample number >=total sample number * sieveing coeffecients a1, increase if being unsatisfactory for
It is loaded this number of days, to obtain enough sample numbers, it is ensured that this part sample is scolded in the first resetting letter area as screening sample
In；
Step (4), carries out the screening sample of the second weight confidence interval, and the second heavy confidence interval bound function of setting carries out
Screening is screened if number of samples is unsatisfactory for expected requirement after adjusting bound function again；If at this time beyond adjustment
Expected requirement is still cannot be satisfied, sample number of days is increased, until sample number meets expected require；
Step (5), extracts the sample filtered out, and active P is multiplied by by normalizedInstallationRenormalization is carried out, is completed
Screening.
Preferably, in the step (1), initialization sample number of days selects 15 days or one week integral multiples.Just
Beginningization sample number of days was 2 to 12 weeks.Initialization sample number of days is rule of thumb typically chosen 15 days, usually 7 days i.e. one week whole
Several times can need to be set as differing in 2 to 8 weeks according to scene, typically not greater than 3 months.In the step (3)：Initialization first
The upper limit function of weight confidence interval:ymax=kmax* x and lower limit function:ymin=kmin* x, the first resetting letter as screening sample
The bound in section, wherein ymin=kmin* x is the first heavy lower limit of confidence interval function；ymax=kmax* x is that area is believed in the first resetting
Between upper limit function；If falling the sample number in the first weight confidence interval<Total sample number * sieveing coeffecients a1, increase sample number of days,
Sample collection is carried out after increasing number of days 2 every time, then is screened；If falling sample number >=sample in the first weight confidence interval
Total * sieveing coeffecients a1, then first again screening finish；Sieveing coeffecient a1Generally between 0.6-0.9.In the step (4)：Into
When the weight confidence interval screening of row second, the upper limit function of initialization the second weight confidence interval：fmaxAnd lower limit function：fmin, with this
Bound of two functions as the second weight confidence interval of screening sample, carries out next step screening sample；fmaxFor the second resetting
Believe section upper limit function, fminFor the second heavy lower limit of confidence interval function, initialization later dynamically adjusts number t=0；If dynamic
State adjusts number t>tmax(dynamic adjustment number maximum value), increases sample number of days, screening sample is carried out after increasing number of days 2 every time；
If dynamic adjustment number t≤tmax, tmaxDynamically to adjust number maximum value, judge to fall the sample in the second weight confidence interval
Whether number is more than or equal to total sample number * sieveing coeffecients a2, if it is, the screening of the second weight confidence interval is completed；If it is not,
Dynamic the second heavy confidence interval upper limit function coefficient k of adjustmentmax=kmax+△kmax, kmaxFor upper limit function coefficient, △ kmaxIt is
Double confidence interval upper limit function coefficient corrects step-length, adjustment upper limit function parameter cmax=cmax+△cmax, cmaxFor upper limit function
Parameter, △ cmaxFor the second heavy confidence interval upper limit function parameters revision step-length；Adjust the second heavy lower limit of confidence interval function coefficients
kmin=kmin-△kmin, kminFor the coefficient of lower limit function, △ kminStep-length is corrected for the second heavy lower limit of confidence interval function coefficients,
Adjust lower limit function parameter cmin=cmin-△cmin, cminThe parameter of lower limit function, △ cminFor the second heavy lower limit of confidence interval letter
Number parameters revision step-length, after often adjustment is primary, dynamic adjusts number t=t+1, then carries out the second heavy screening sample, until falling
Sample number >=total sample number * sieveing coeffecients a in second weight confidence interval2, then the sample sieve in the second weight confidence interval is completed
Choosing.
1) days=15 days wind speed and active sample are chosen in initialization；
2) air speed value remains unchanged in sample, active PIt is real/PInstallationIt is normalized, establishes two-dimensional coordinate system (wherein PIt is realIt is wind
The reality of electric field is active, PInstallationIt is the installed capacity of wind power plant)；
3) upper limit function of the first weight of initialization confidence interval:ymax=k1max* x and lower limit function:ymin=k1min* x makees
For the bound of the first weight confidence interval of screening sample
(wherein k1maxFor the coefficient of the first heavy confidence interval upper limit function, k1minFor the first heavy lower limit of confidence interval function
Coefficient)；
If 4) fall the sample number in the first weight confidence interval<Total sample number * sieveing coeffecients a1, increase sample number of days,
After increasing number of days 2 every time, sample collection is carried out, enters step and 2) continues to execute；
If 5) fall sample number >=total sample number * sieveing coeffecients a in the first weight confidence interval1, sieved again into second
Choosing；
6) second again screening start, initialization second weight confidence interval upper limit function:fmaxAnd lower limit function:fmin, with
Bound of the two functions as the second weight confidence interval of screening sample, carries out next step screening sample；
7) initialization dynamic adjustment number t=0；
If 8) dynamic adjustment number t>tmax(dynamic adjustment number maximum value), increases sample number of days, increases number of days every time
Sample collection is carried out after 2, is entered step and 2) is continued to execute；
If 9) dynamic adjustment number t≤tmax10) and the sieve of step 11) (dynamic adjustment number maximum value) enters step
Choosing；
If 10) fall the sample number in the second weight confidence interval<Total sample number * sieveing coeffecients a2, dynamic adjustment second
Weight confidence interval upper limit function coefficient kmax=kmax+△kmax, adjustment upper limit function parameter cmax=cmax+△cmax；Adjustment second
Weight lower limit of confidence interval function coefficients kmin=kmin-△kmin, adjustment lower limit function parameter cmin=cmin-△cmin, often adjust one
After secondary, dynamic adjusts number t=t+1；It enters step 8)；
(wherein, △ kmaxStep-length, △ c are corrected for the second heavy confidence interval upper limit function coefficientmaxFor the second heavy confidence interval
Upper limit function parameters revision step-length, △ kminStep-length, △ c are corrected for the second heavy lower limit of confidence interval function coefficientsminFor the second weight
Lower limit of confidence interval function parameter corrects step-length)；
If 11) fall sample number >=total sample number * sieveing coeffecients a in the second weight confidence interval2, sample is filtered out,
Active it is multiplied by P by normalizedInstallationRenormalization is carried out, screening is completed.
Claims (5)
1. the wind farm power prediction screening sample method based on dual confidence interval, it is characterised in that include the following steps：
Step (1), initialization sample select number of days；
Step (2), it is active to reality to be normalized, establishes two-dimensional coordinate system so that the reality after wind speed and normalization is active；
Step (3), initialization first weigh the bound function of confidence interval, and first as screening sample weighs the upper of confidence interval
Lower limit, it is ensured that fall sample number >=total sample number * sieveing coeffecient a1 in the first weight confidence interval, increase sample if being unsatisfactory for
This number of days, to obtain enough sample numbers, it is ensured that this part sample is scolded in the first weight confidence interval as screening sample；
Step (4), carries out the screening sample of the second weight confidence interval, and the second heavy confidence interval bound function of setting is sieved
Choosing is screened if number of samples is unsatisfactory for expected requirement after adjusting bound function again；If beyond adjustment at this time according to
It is old to cannot be satisfied expected requirement, increase sample number of days, until sample number meets expected require；
Step (5), extracts the sample filtered out, is multiplied by P by the reality after normalization is activeInstallationRenormalization is carried out, it is complete
At screening.
2. the wind farm power prediction screening sample method based on dual confidence interval as described in claim 1, it is characterized in that：Institute
It states in step (1), initialization sample number of days selects 15 days or one week integral multiples.
3. the wind farm power prediction screening sample method based on dual confidence interval as claimed in claim 2, it is characterized in that：It can
Initialization sample number of days was 2 to 12 weeks.
4. the wind farm power prediction screening sample method based on dual confidence interval as described in claim 1, it is characterized in that：Institute
It states in step (3), the upper limit function of initialization the first weight confidence interval:Ymax=kmax*x and lower limit function:Ymin=kmin*
X, the bound of the first weight confidence interval as screening sample, wherein ymin=kmin*x is the first heavy lower limit of confidence interval letter
Number；Ymax=kmax*x is the first heavy confidence interval upper limit function；If falling the sample number in the first weight confidence interval<Sample
Total * sieveing coeffecients a1 increases sample number of days, carries out sample collection after increasing number of days 2 every time, then screened；If fallen
First weight confidence interval in sample number >=total sample number * sieveing coeffecient a1, then first again screening finish；Sieveing coeffecient a1 exists
Between 0.6-0.9.
5. the wind farm power prediction screening sample method based on dual confidence interval as described in claim 1, it is characterized in that：Institute
It states in step (4)：
When carrying out the second weight confidence interval screening, the upper limit function of initialization the second weight confidence interval：Fmax and lower limit function：
Fmin carries out next step screening sample using the two functions as the bound of the second of screening sample the weight confidence interval；fmax
For the second heavy confidence interval upper limit function, fmin is the second heavy lower limit of confidence interval function, later initialization dynamic adjustment number t
=0；If dynamic adjustment number t>Tmax, tmax are dynamic adjustment number maximum value, increase sample number of days, increase number of days every time
Screening sample is carried out after 2；If dynamic adjustment number t≤tmax, whether judgement falls the sample number in the second weight confidence interval
More than or equal to total sample number * sieveing coeffecient a2, if it is, the screening of the second weight confidence interval is completed；If it is not, dynamic is adjusted
Whole second heavy confidence interval upper limit function coefficient k max=kmax+ △ kmax, kmax are upper limit function coefficient, and △ kmax are second
Weight confidence interval upper limit function coefficient corrects step-length, and adjustment upper limit function parameter cmax=cmax+ △ cmax, cmax are upper limit letter
Number parameter, △ cmax are the second heavy confidence interval upper limit function parameters revision step-length；Adjust the second heavy lower limit of confidence interval function
Coefficient k min=kmin- △ kmin, kmin are the coefficient of lower limit function, and △ kmin are the second heavy lower limit of confidence interval function coefficients
Step-length is corrected, the parameter of lower limit function parameter cmin=cmin- △ cmin, cmin lower limit functions is adjusted, △ cmin are the second weight
Lower limit of confidence interval function parameter corrects step-length, and after often adjustment is primary, dynamic adjusts number t=t+1, then carries out second same
Screening then completes the second resetting letter area until falling sample number >=total sample number * sieveing coeffecient a2 in the second weight confidence interval
Interior screening sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510729241.4A CN106650191B (en) | 2015-10-31 | 2015-10-31 | Wind farm power prediction screening sample method based on dual confidence interval |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510729241.4A CN106650191B (en) | 2015-10-31 | 2015-10-31 | Wind farm power prediction screening sample method based on dual confidence interval |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106650191A CN106650191A (en) | 2017-05-10 |
| CN106650191B true CN106650191B (en) | 2018-11-13 |
Family
ID=58810461
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510729241.4A Active CN106650191B (en) | 2015-10-31 | 2015-10-31 | Wind farm power prediction screening sample method based on dual confidence interval |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106650191B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184344A (en) * | 2011-06-17 | 2011-09-14 | 上海电机学院 | Method and device for determining confidence probability of power prediction result of wind power station |
| CN103296701A (en) * | 2013-05-09 | 2013-09-11 | 国家电网公司 | Active power control method in wind power plant |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6003247B2 (en) * | 2012-06-05 | 2016-10-05 | 富士電機株式会社 | Information processing apparatus, information processing apparatus control method, and program |
-
2015
- 2015-10-31 CN CN201510729241.4A patent/CN106650191B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184344A (en) * | 2011-06-17 | 2011-09-14 | 上海电机学院 | Method and device for determining confidence probability of power prediction result of wind power station |
| CN103296701A (en) * | 2013-05-09 | 2013-09-11 | 国家电网公司 | Active power control method in wind power plant |
Non-Patent Citations (2)
| Title |
|---|
| Creating Confidence Intervals for Reservoir Computing’s Wind Power Forecast Use of Maximum Likelihood Method and the Distribution-based Method;Breno Menezes et al.;《COGNITIVE 2014 : The Sixth International Conference on Advanced Cognitive Technologies and Applications》;20141231;第172-177页 * |
| 基于OS-ELM和Bootstrap方法的超短期风电功率预测;王焱 等;《电力系统自动化》;20140325;第38卷(第6期);第14-19、122页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106650191A (en) | 2017-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021073462A1 (en) | 10 kv static load model parameter identification method based on similar daily load curves | |
| CN104036474B (en) | A kind of Automatic adjustment method of brightness of image and contrast | |
| CN103971342B (en) | A kind of image noise detection method based on convolutional neural networks | |
| US20210109140A1 (en) | Method for identifying parameters of 10 kv static load model based on similar daily load curves | |
| CN103494316B (en) | A kind of method dividing redried leaf tobacco batch by weight | |
| CN106251242B (en) | Wind power output interval combination prediction method | |
| CN109440515A (en) | Soft measurement method for beating degree in papermaking process based on gradient enhanced regression tree | |
| CN109344738A (en) | The recognition methods of crop diseases and pest crop smothering and device | |
| CN112330050A (en) | Power system load prediction method considering multiple features based on double-layer XGboost | |
| CN108038616A (en) | A kind of work transmission line analysis of prices method | |
| CN106355338A (en) | Raw milk risk detection and control method | |
| CN116562623A (en) | Real-time planting risk assessment system suitable for velvet mushroom is planted | |
| CN104077489A (en) | Method and system for analyzing energy efficiency of energy consumption device | |
| CN106650191B (en) | Wind farm power prediction screening sample method based on dual confidence interval | |
| CN111932391A (en) | Calculation method for particle size distribution data consistency correction in ore crushing or grinding process investigation | |
| Chalfant et al. | Marketing orders, grading errors, and price discrimination | |
| Wiles et al. | Modeling competition for light between soybean and broadleaf weeds | |
| CN105960587A (en) | Method and system for monitoring the quality of photovoltaic cells | |
| CN113313287B (en) | Construction method of short-term prediction model of population quantity of Laodelphax striatellus | |
| CN110426669A (en) | A kind of automatic screening methods of electric energy meter based on error testing | |
| CN110456356B (en) | Data assimilation-oriented weather radar radial speed quality control method | |
| CN106655278A (en) | Operation data-based wind power farm active power and reactive voltage control evaluation method | |
| CN106719284A (en) | A kind of sea-farming wisdom weather forecast service system | |
| CN116485517A (en) | Intelligent credit service management system | |
| CN110298762A (en) | A kind of family ranch culture efficiency analysis system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |