CN104615855A - Day-ahead wind speed multistep prediction method fused with numerical weather prediction - Google Patents
Day-ahead wind speed multistep prediction method fused with numerical weather prediction Download PDFInfo
- Publication number
- CN104615855A CN104615855A CN201510004112.9A CN201510004112A CN104615855A CN 104615855 A CN104615855 A CN 104615855A CN 201510004112 A CN201510004112 A CN 201510004112A CN 104615855 A CN104615855 A CN 104615855A
- Authority
- CN
- China
- Prior art keywords
- forecast
- wind speed
- numerical weather
- model
- result
- 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
Abstract
The invention provides a day-ahead wind speed multistep prediction method fused with numerical weather prediction, and belongs to the technical field of wind speed prediction. The problems that an existing method is directly fused with numerical weather prediction, the prediction error is large, and the prediction precision of wind electricity power is low are solved. According to the technical scheme, the method comprises the steps of analyzing the information effectiveness of an original wind velocity sequence, determining the predictability time duration of a statistic forecasting model, obtaining the wind speed prediction result within a predicable time duration, conducting information effectiveness analysis on a numerical weather prediction model, determining the predicable time duration of the numerical weather prediction model, obtaining the wind speed prediction result within the predicable time duration, establishing a day-ahead wind speed fusion prediction model according to the wind speed prediction result and predicting the actual wind speed. The day-ahead wind speed multistep prediction method fused with numerical weather prediction is suitable for predicting the wind speed within day-ahead 24 hours in the future.
Description
Technical field
The present invention relates to a kind of multi-step prediction method of novel wind speed a few days ago, particularly one merges the multi-step prediction method of wind speed a few days ago of numerical weather forecast (NWP), belongs to wind speed forecasting technique field.
Background technology
Wind-power electricity generation is the hot issue of current new forms of energy power industry research.But because wind has very strong intermittence and uncontrollability, therefore wind power is also to fluctuate with the wind and uncontrollable.This uncontrollability changes the original operational mode of electrical network (being filled with strong stochastic uncertainty power supply at Generation Side), such that large-scale wind power is grid-connected causes serious impact afterwards to electrical network, affects the safe and stable operation of electrical network.Wind energy turbine set wind power is predicted accurately to be the important foundation solving this problem, be conducive to dispatching of power netwoks department and grasp wind power output situation in time, adjust operation plan in time, determine spinning reserve.And the wind speed topmost influence factor that is wind power, therefore carrying out Accurate Prediction to wind speed, is the important prerequisite of wind power prediction.Be wherein that the forecast of wind speed a few days ago of 1-24 hours of unit can help electrical network rational management with " hour ", ensure power supply quality, for wind energy turbine set participates in surfing the Net at a competitive price and giving security.
In recent years, lot of domestic and international scholar starts to pay close attention to the wind speed forecasting method a few days ago merging numerical weather forecast, and expands relevant research.Wherein, the thinking of being used widely utilizes the forecast result of statistical model (as neural network etc.) logarithm value weather forecast to revise, thus obtain final forecast result.A.Vaccaroa uses mesoscale numerical weather forecast, partial high-precision numerical weather forecast, actual measurement wind speed and meteorologic parameter to form a proper vector, devise the Lazy learning algorithm based on K next-door neighbour, predict with the mean value of k in history most similarity vector.Federico employs the way that NWP is combined with Kalman filtering, and dynamic corrections is carried out in predicting the outcome of logarithm value weather forecast, points out that Kalman filtering algorithm can reduce the systematic error in NWP.The wind speed that Cai Zhenqi forecasts with history NWP and related data are input, and corresponding time period measured data, for exporting neural network training, obtains correction model, subsequently with the NWP data of forecast day for input, obtain the NWP wind speed of forecast day correction.But the validity problem of numerical weather forecast and these two kinds of information of measured data is not considered in current research, but directly simply merge after predicting same time length respectively by two kinds of methods.In fact, the predictable period of numerical weather forecast and statistical method also exists very large difference.Actual measurement air speed data is only containing short period transient component, and not containing the meteorological process component of long period, so the forecasting wind speed value utilizing statistical method to obtain is with a high credibility in short-term, and when predicted time increases, the precision of forecast can reduce.Numerical weather forecast (NWP) is according to local weather conditions, time equal meaning under carry out Closure equation group with turbulence model, and air Basic equation group is solved under starting condition and boundary condition, the atmospheric condition of forecast future time instance, obtain forecasting wind speed value, so the Output rusults of NWP system is the room and time mean value of each computing grid, and does not simulate the transient process of turbulent flow.Therefore the Computing Principle of NWP determines the prediction ability of NWP, and it is containing the meteorological process component of long period, and not containing short period turbulent flow component, it is 1h that result of calculation exports step-length, and the effect of instantaneous forecast is undesirable.
Therefore, Water demand numerical weather forecast and statistical method predictable period separately, then takes certain fusion method to set up forecasting model according to respective effectiveness of information and provides forecast result.
Summary of the invention
The object of the invention is to propose a kind of multi-step prediction method of wind speed a few days ago merging numerical weather forecast, large to solve the predicated error merged for the direct numerical weather forecast of existing method, the problem that the precision of prediction of wind power is low.
The present invention for solving the problems of the technologies described above adopted technical scheme is:
The multi-step prediction method of wind speed a few days ago of fusion numerical weather forecast of the present invention, realizes according to following steps:
Step one, analyze the effectiveness of information of original wind series, determine the forecast duration of statistical forecast model, and obtain the wind speed forecast result that can forecast in duration;
Step 2, logarithm value Forecast Model For Weather carry out analyzing information validity, determine the forecast duration of numerical weather forecast model, and obtain the wind speed forecast result that can forecast in duration;
Step 3, the wind speed forecast result obtained according to step one and step 2, set up wind speed a few days ago and merge forecasting model, forecast actual wind speed.
The invention has the beneficial effects as follows:
1, the present invention analyzes the effectiveness of information of numerical weather forecast and measured data, obtains otherness and the complementarity of the forecast result of two kinds of methods: numerical weather forecast only containing the meteorological process component of long period, and does not contain short period turbulent flow component; And utilize the prediction effect of the statistical method short-term of measured data better, along with the forecast of growth effect of predicted time can be deteriorated.So, two kinds of methods advantage is separately utilized to carry out complementation, different amalgamation modes is adopted according to the validity of two kinds of information within the different forecast time periods, obtain final forecast result, compared to the statistical fluctuation result not merging numerical weather forecast, in 24 hours, the average relative error of forecast result reduces all to some extent, maximumly reduces 25%, thus improve the precision of wind speed forecast, meet wind-electricity integration requirement.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of the inventive method;
Fig. 2 is Mallat wavelet decomposition algorithm schematic diagram;
Fig. 3 is Mallat wavelet reconstruction algorithm schematic diagram;
Fig. 4 is the air speed data figure of actual measurement;
Fig. 5 is three layers of Multiscale Wavelet Decomposition result figure of air speed data, the third layer low frequency component obtained after being followed successively by three layers of wavelet decomposition in figure from top to bottom
third layer high fdrequency component
second layer high fdrequency component
and ground floor high fdrequency component
Fig. 6 be the auto-correlation function value of each dimensions in frequency vector sequence with the change curve of persistence length, be corresponding in turn to from top to bottom in figure
and
the autocorrelation analysis result of four components;
The persistence length that Fig. 7 coefficient of autocorrelation reaches 0.8 is along with the change curve of the wavelet decomposition number of plies;
Fig. 8 is the change curve of the average relative error that predicts the outcome of difference with prediction step;
Fig. 9 is the change curve of the root-mean-square error that predicts the outcome of difference with prediction step;
Figure 10 is numerical weather forecast error map;
What Figure 11 obtained for the multi-step prediction method of wind speed a few days ago merging numerical weather forecast (NWP) predicts the outcome and the comparison diagram of surveying wind speed;
Figure 12 be on test set average relative error with the change curve of prediction step.
Embodiment
Embodiment one: composition graphs 1 illustrates present embodiment, a kind of multi-step prediction method of wind speed a few days ago merging numerical weather forecast described in present embodiment, comprises the following steps:
Step one, analyze the effectiveness of information of original wind series, determine the forecast duration of statistical forecast model, and obtain the wind speed forecast result that can forecast in duration;
Step 2, logarithm value Forecast Model For Weather carry out analyzing information validity, determine the forecast duration of numerical weather forecast model, and obtain the wind speed forecast result that can forecast in duration;
Step 3, the wind speed forecast result obtained according to step one and step 2, carry out integration modeling according to the otherness of numerical weather forecast model and measured data and complementarity, set up wind speed a few days ago and merge forecasting model, forecast actual wind speed.
The thinking of the multi-step prediction method of wind speed a few days ago of the fusion numerical weather forecast (NWP) described in present embodiment is as follows: when we recognize that the forecast result that numerical weather forecast (NWP) obtains is under equal meaning, it is containing the meteorological process component of long period, and not containing short period turbulent flow component; And survey air speed data only containing short period transient component, not containing the meteorological process component of long period, so the prediction step of statistical method is subject to certain limitation.Under these two prerequisites, it is inappropriate that numerical weather forecast and statistical method directly being merged in identical predicted time in the past carries out wind speed forecast a few days ago.
The beneficial effect of present embodiment is:
1, analyze the effectiveness of information of numerical weather forecast and measured data, obtain otherness and the complementarity of the forecast result of two kinds of methods: numerical weather forecast only containing the meteorological process component of long period, and does not contain short period turbulent flow component; And utilize the prediction effect of the statistical method short-term of measured data better, along with the forecast of growth effect of predicted time can be deteriorated.So, two kinds of methods advantage is separately utilized to carry out complementation, different amalgamation modes is adopted according to the validity of two kinds of information within the different forecast time periods, obtain final forecast result, compared to the statistical fluctuation result not merging numerical weather forecast, in 24 hours, the average relative error of forecast result reduces all to some extent, maximumly reduces 25%, thus improve the precision of wind speed forecast, meet wind-electricity integration requirement.
Embodiment two: composition graphs 2, Fig. 3 illustrate present embodiment, present embodiment and embodiment one, unlike the effectiveness of information of the original wind series of the analysis described in step one, determine that the process of the forecast duration of statistical forecast model is:
Step one by one, before utilizing statistical forecast model, the multiple dimensioned characteristic of consideration wind series, utilizes wavelet decomposition algorithm to decompose original wind series;
Step one two, the subsequence of each frequency that obtains after wavelet decomposition in original wind series carry out autocorrelation analysis, determines the forecast duration on each frequency subsequence.
Embodiment three: composition graphs 2, Fig. 3, Fig. 4 illustrate present embodiment, present embodiment and embodiment one or two unlike: the detailed process of the wind speed forecast result that obtaining described in step one can be forecast in duration is:
Step one three, on the basis of step one two, each frequency subsequence utilizes statistical forecast model (as SVR model) respectively, the model of each subsequence only forecasts that this subsequence can forecast the wind speed within the scope of duration.Obtain each frequency subsequence can forecast and wind speed forecast result in duration synthesize, obtain the wind speed forecast result of statistical forecast model to the forecast result of the statistical forecast model of each subsequence.
Embodiment four: composition graphs 2, Fig. 3 illustrate present embodiment, one of present embodiment and embodiment one to three unlike the wavelet decomposition algorithm that utilizes that: step is described one by one to the detailed process that original wind series is decomposed are:
Because the factor (as temperature, air pressure, roughness of ground surface, general circulation etc.) affecting wind speed is numerous, the mechanism of action is complicated, very strong multiple dimensioned characteristic is revealed in the sequence table of wind series sequence nucleotide sequence, namely the sequence nucleotide sequence sequence nucleotide sequence frequency that different effect sources produces is different, and therefore final wind series can regard the result that multiple sequence nucleotide sequence sequence nucleotide sequence is coupled as.So the present invention is before setting up statistical forecast model,
Mallat wavelet algorithm is utilized to carry out three layers of wavelet decomposition to original wind series.
Multiresolution analysis and Mallat algorithm:
Space L
2(R) multiresolution analysis on refers to one group of subspace by constructing in this space
make it have following character:
(1) monotonicity: subspace also exists strict relation of inclusion
(2) Approximation:
(3) retractility:
(4) translation invariance:
(5) existence: exist
make
form V
jriesz base.
Theorem: if
l
2(R) multiresolution analysis on, then the function of existence anduniquess
make
for V
jan interior orthonormal basis, wherein
be called as scaling function.
The concept introducing scaling function is to construct orthogonal wavelet function.If
generate multiresolution analysis
Simultaneously
v
-1on standard Riesz base, then
can be by
represent, that is:
Monotonicity can by V
juse V
j+1represent with orthocomplement, that is:
V
j＝V
j+1⊕W
j+1， (6)
Get the limit can obtain,
another ψ (t) generate space W
0, ψ
j,k(t) generate space W
j, { W
j}
j ∈ Zbe mutually orthogonal sequence of subspaces, be otherwise known as wavelet space.ψ (t) can by V again
-1on basis representation
Above equation is the Double-scaling equation of scaling function and wavelet function, the structure of wavelet function and scaling function, can be summed up as coefficient { g (k) }
k ∈ Z{ h (k) }
k ∈ Zdesign.Order
So the design of scaling function and wavelet function can be converted into filters H (ω), the design of G (ω).
Mallat, under the inspiration of pyramid algorith, in conjunction with multiresolution analysis, gives the pyramid wavelet function feedback algorithm of sequence nucleotide sequence sequence nucleotide sequence.If sequence nucleotide sequence sequence nucleotide sequence f (t) is at metric space V
jwith wavelet space W
jbe projected as
Might as well establish
d
j,k=<f (t), ψ
j,kt () >, by V
j=V
j+1⊕ W
j+1, can obtain
Can be obtained by the Double-scaling equation of scaling function
Can be obtained by scaling function orthogonality
Can be obtained by wavelet function Double-scaling equation
It is more than simultaneous that three equations can obtain:
Above formula is Mallat decomposition and reconstruction algorithm, and schematic diagram as shown in Figures 2 and 3;
By using Mallat wavelet decomposition algorithm, to certain wind energy turbine set in May, 2011 to three months July, decompose by the air speed datas of 10 minutes (shown in Fig. 4).Fig. 5 is the result that original wind speed obtains after three layers of wavelet decomposition.After original wind series being decomposed into the subsequence of different frequency, just can set up statistical forecast model for each subsequence.Embodiment five: Fig. 6, Fig. 7, Fig. 8 illustrate present embodiment, one of present embodiment and embodiment one to four unlike: the detailed process of the forecast duration on each frequency subsequence of the determination described in step one two is:
Convenient in order to describe, this by original wind series through three layers of wavelet decomposition, the high fdrequency component of the ground floor obtained is designated as
the high fdrequency component of the second layer is designated as
the high fdrequency component of third layer is designated as
the low frequency component of third layer is designated as
The classical way investigating time series autocorrelation is Pearson auto-relativity function method, if { x
t}
t=1:nbe a Random time sequence, then measure x
twith the sample x of its delay k step-length
t+kcoefficient of autocorrelation be defined as the covariance of sample, that is:
Autocorrelation function is defined as
according to the definition of autocorrelation function, calculate the auto-correlation function value of each frequency component, with persistence length change curve as shown in Figure 6.What Fig. 7 represented is that after wavelet decomposition, each layer air speed data coefficient of autocorrelation reaches the persistence length of 0.8.
When auto-correlation function value is more than or equal to 0.8, very strong association is there is between data, forecast result confidence level in persistence length corresponding when can think that auto-correlation function value is greater than 0.8 is higher, and namely corresponding persistence length is to forecast duration, and the credible result degree namely forecast is higher.
Auto-correlation function value lower than 0.8 time, the correlativity between data is strong, and the confidence level of the wind speed forecast result in corresponding persistence length is lower, so take from correlation function value to be more than or equal to the persistence length of more than 0.8 as forecasting duration.
The low frequency component of third layer is designated as
observe auto-correlation function value with the change curve of persistence length, can find out that auto-correlation function value starts lower than 0.8 after persistence length was more than 4 hours, and along with the increase of persistence length, auto-correlation function value reduces always.Therefore the forecast duration of low frequency component is got 4 hours is comparatively suitable.In like manner, study other each frequency components,
upper getting respectively can forecast that duration is 1 hour, 0.4 hour, 0.2 hour.Therefore can find out that its autocorrelation of subsequence of different frequency range is different, low frequency component tendency is strong, and its time that can forecast is longer; High fdrequency component randomness is strong, and its time that can forecast is shorter.
For the statistical forecast model that each frequency component is set up, the auto-correlation length of each component multi-step prediction is taken as 4 hours respectively, 1 hour, 0.4 hour, 0.2 hour.The input of the statistical forecast model of every one deck is chosen according to auto-correlation length equally, and the air speed value of namely getting current time and the moment of L-1 before forms input vector, and wherein L is the numerical value of this layer of auto-correlation length.
After obtaining the forecast result of each layer, utilize Mallat wavelet reconstruction algorithm to carry out multiple dimensioned synthesis to the forecast result of each layer, the predictable period namely on every one deck is different, is synthesized by the forecast result on each yardstick according to respective calling time in advance.Each layer component only forecasts its time that can forecast, for high fdrequency component, it is responsible for the forecast of following 0.2 hour, other frequency components the like.Time be greater than 1 hour when calling time in advance after, its forecast result is only that the forecast result of low frequency component provided.
Can be found out by the analysis of step one, adopt statistical forecast model to carry out wind speed a few days ago and give the correct time in advance, the wind speed forecast within following 4 hours is with a high credibility, and beyond 4 hours, the confidence level of forecast result starts to reduce.
Embodiment six: composition graphs 8, Fig. 9, Figure 10 illustrate present embodiment, one of present embodiment and embodiment one to five unlike: described in step 2, the detailed process of the forecast duration of fixed number value Forecast Model For Weather is really:
According to the ultimate principle of numerical weather forecast and the analysis of forecast result: with time equal meaning under the physical law such as the mass conservation, momentum conservation, energy conservation and steam conservation set up air Basic equation group, and carry out Closure equation group with turbulence model, atmospheric condition is solved, the atmospheric condition of forecast future time instance under starting condition and boundary condition.Generally adopt typical WRFa Forecast Mode at present, adopt three layers of nested grid, computing grid is that level is respectively: 27km × 27km, 9km × 9km, 3km × 3km.The Output rusults of NWP system is the room and time mean value of each computing grid, does not simulate the transient process of turbulent flow.So the Computing Principle of NWP determines the prediction ability of NWP, it is containing the meteorological process component of long period, and not containing short period turbulent flow component, it is 1h that result of calculation exports step-length, and the result of instantaneous forecast is poor.
When statistical forecast model (as SVM model) and numerical weather forecast (NWP) forecast wind speed, average relative error and the root-mean-square error curve of forecast result are analyzed.Can find out when calling time in advance within 4 hours, the forecast result of numerical weather forecast (NWP) and the forecast result of SVM are more or less the same, but once call time in advance after more than 4 hours, no matter be average relative error or root-mean-square error, the forecast result of numerical weather forecast (NWP) all will be better than the forecast result of SVM far away.Figure 10 is the distribution of numerical weather forecast error.
The analytic process of described forecast result is as follows:
Fig. 8 and Fig. 9 utilizes statistical forecast model (as SVM model) and numerical weather forecast model (NWP) to forecast wind speed, the average relative error of comparison forecast result and root-mean-square error are with the change curve increased that calls time in advance, two width curve maps are found out, draw when calling time in advance within 4 hours, the forecast result of the forecast result of numerical weather forecast model (NWP) and statistical forecast model (as SVM model) is more or less the same, call time in advance after more than 4 hours, no matter be average relative error or root-mean-square error, the forecast result of numerical weather forecast model (NWP) all will be better than the forecast result of statistical forecast model (as SVM model) far away.Figure 10 is the distribution of numerical weather forecast error.
The forecast result of 4 ~ 24 h values Forecast Model For Weathers (NWP) is higher than the precision of statistical forecast model (as SVM model) forecast result, show that the forecast of 4 ~ 24 hours is realized by numerical weather forecast model (NWP), thus determine that the forecast duration of numerical weather forecast model (NWP) is 4 ~ 24 hours.Can be found out by the analysis of logarithm value weather forecast information validity, the prediction ability of NWP is comparatively strong, and the long value of forecasting is better, and the instantaneous value of forecasting is poor.
Embodiment seven: unlike: the foundation described in the step 3 detailed process that wind speed merges forecasting model be a few days ago in conjunction with one of Figure 11, Figure 12 present embodiment and embodiment one to six:
As can be seen from the analyzing information validity of numerical weather forecast and measured data, utilize statistical model to carry out wind speed to give the correct time in advance, when calling time in advance after more than 4 hours, the autocorrelation functional value of low frequency component is lower than 0.8, and reduce along with the increase of calling time in advance, therefore the confidence level of forecast result starts to reduce.And the prediction ability of numerical weather forecast (NWP) is comparatively strong, but the instantaneous value of forecasting is poor.So the thinking that statistical model and numerical weather forecast (NWP) carry out merging is that within 4 hours, wind speed forecast can obtain effective information from numerical weather forecast (NWP) and historical wind speed by the present invention,
Adopt genetic algorithm the forecast result of numerical weather forecast model and the forecast result of statistical forecast model to be weighted, obtain the wind speed forecast result within 4 hours; The wind speed forecast result of 4 ~ 24 hours is obtained by numerical weather forecast model (NWP).
Merge forecasting model as follows:
v＝w
1v
1+w
2v
2
In formula, v is the final forecast result of wind speed a few days ago, v
1, v
2be respectively the wind speed forecast result of statistical forecast model and numerical weather forecast model; w
1, w
2for v time diffusion-weighted
1, v
2corresponding weights, when wherein synthesizing in 4 hours, w
1, w
2obtained by genetic algorithm, and w during the synthesis of 4 ~ 24 hours
1=0, w
2=1, the wind speed forecast result namely in 4 hours is obtained by the forecast result synthesis of statistical forecast model and numerical weather forecast model, and the wind speed forecast result of 4 ~ 24 hours is only obtained by numerical weather forecast model (NWP).
Figure 11 is the comparison diagram merging the curve of wind series a few days ago that obtains of the multi-step prediction method of wind speed a few days ago of numerical weather forecast (NWP) and the wind series curve of actual measurement.Can find out, forecast wind speed is higher with the goodness of fit of actual measurement wind speed, demonstrates validity of the present invention.
In order to illustrate further beneficial effect of the present invention.The value of forecasting of the multi-step prediction method of wind speed a few days ago that same test set compares the fusion numerical weather forecast (NWP) that the present invention proposes and the statistical method not merging NWP.The average relative error of different forecasting procedure is along with forecasting the change curve of step-length as shown in figure 12.By relatively finding out, within the scope of whole test set, along with the increase of forecast step-length, the value of forecasting that the wind speed multi-step prediction method merging numerical weather forecast (NWP) obtains all will be better than the value of forecasting of the statistical method gained not merging NWP greatly, again demonstrates validity of the present invention.
Claims (7)
1. merge the multi-step prediction method of wind speed a few days ago of numerical weather forecast, it is characterized in that said method comprising the steps of:
Step one, analyze the effectiveness of information of original wind series, determine the forecast duration of statistical forecast model, and obtain the wind speed forecast result that can forecast in duration;
Step 2, logarithm value Forecast Model For Weather carry out analyzing information validity, determine the forecast duration of numerical weather forecast model, and obtain the wind speed forecast result that can forecast in duration;
Step 3, the wind speed forecast result obtained according to step one and step 2, set up wind speed a few days ago and merge forecasting model, forecast actual wind speed.
2. the multi-step prediction method of wind speed a few days ago of fusion numerical weather forecast according to claim 1, is characterized in that the effectiveness of information of the original wind series of analysis described in step one, determines that the process of the forecast duration of statistical forecast model is:
Step one by one, before utilizing statistical forecast model, utilize wavelet decomposition algorithm to decompose original wind series;
Step one two, the subsequence of each frequency that obtains after wavelet decomposition in original wind series carry out autocorrelation analysis, determines the forecast duration on each frequency subsequence.
3. the multi-step prediction method of wind speed a few days ago of fusion numerical weather forecast according to claim 2, is characterized in that the detailed process of the wind speed forecast result that obtaining described in step one can be forecast in duration is:
Step one three, on the basis of step one two, each frequency subsequence utilizes statistical forecast model respectively, obtain each frequency subsequence and can forecast wind speed forecast result in duration, the forecast result of the statistical forecast model of each subsequence is synthesized, obtains the wind speed forecast result of statistical forecast model.
4. the multi-step prediction method of wind speed a few days ago of fusion numerical weather forecast according to claim 3, is characterized in that the wavelet decomposition algorithm that utilizes described in step is one by one to the detailed process that original wind series is decomposed is:
Mallat wavelet algorithm is utilized to carry out three layers of wavelet decomposition to original wind series.
5. the multi-step prediction method of wind speed a few days ago of fusion numerical weather forecast according to claim 4, is characterized in that the detailed process of the forecast duration on each frequency subsequence of determination described in step one two is:
By original wind series through three layers of wavelet decomposition, according to the definition of autocorrelation function, calculate the auto-correlation function value of each frequency component, when auto-correlation function value is more than or equal to 0.8, corresponding persistence length to forecast duration.
6. the multi-step prediction method of wind speed a few days ago of fusion numerical weather forecast according to claim 5, is characterized in that the detailed process of the forecast duration of fixed number value Forecast Model For Weather really described in step 2 is:
Utilize statistical forecast model and numerical weather forecast model prediction wind speed, compare the average relative error of forecast result and root-mean-square error with the change curve increased that calls time in advance, determine that the forecast duration of numerical weather forecast model is 4 ~ 24 hours.
7. the multi-step prediction method of wind speed a few days ago of fusion numerical weather forecast according to claim 6, is characterized in that the detailed process of the foundation wind speed fusion a few days ago forecasting model described in step 3 is:
Adopt genetic algorithm the forecast result of numerical weather forecast model and the forecast result of statistical forecast model to be weighted, merge forecasting model as follows:
v＝w
1v
1+w
2v
2
In formula, v is the final forecast result of wind speed a few days ago, v
1, v
2be respectively the wind speed forecast result of statistical forecast model and numerical weather forecast model; w
1, w
2for v time diffusion-weighted
1, v
2corresponding weights, when wherein synthesizing in 4 hours, w
1, w
2obtained by genetic algorithm, and w during the synthesis of 4 ~ 24 hours
1=0, w
2=1, the wind speed forecast result namely in 4 hours is obtained by the forecast result synthesis of statistical forecast model and numerical weather forecast model, and the wind speed forecast result of 4 ~ 24 hours is only obtained by numerical weather forecast model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510004112.9A CN104615855A (en) | 2015-01-04 | 2015-01-04 | Day-ahead wind speed multistep prediction method fused with numerical weather prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510004112.9A CN104615855A (en) | 2015-01-04 | 2015-01-04 | Day-ahead wind speed multistep prediction method fused with numerical weather prediction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104615855A true CN104615855A (en) | 2015-05-13 |
Family
ID=53150296
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510004112.9A Pending CN104615855A (en) | 2015-01-04 | 2015-01-04 | Day-ahead wind speed multistep prediction method fused with numerical weather prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104615855A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105576650A (en) * | 2015-12-29 | 2016-05-11 | 北京国能日新系统控制技术有限公司 | Wind power prediction method and system based on assimilation technology and same wind band theory |
| CN105956252A (en) * | 2016-04-27 | 2016-09-21 | 哈尔滨工业大学 | Generative deep belief network-based multi-scale forecast modeling method for ultra-short-term wind speed |
| CN106505631A (en) * | 2016-10-29 | 2017-03-15 | 塞壬智能科技（北京）有限公司 | Intelligent wind power wind power prediction system |
| CN108595658A (en) * | 2018-04-28 | 2018-09-28 | 尚谷科技（天津）有限公司 | A kind of weather knowledge base detection method based on multimedia content |
| CN109214575A (en) * | 2018-09-12 | 2019-01-15 | 河海大学 | A kind of super short-period wind power prediction technique based on small wavelength short-term memory network |
| CN111008604A (en) * | 2019-12-09 | 2020-04-14 | 上海眼控科技股份有限公司 | Prediction image acquisition method and device, computer equipment and storage medium |
| CN111339092A (en) * | 2020-02-24 | 2020-06-26 | 江苏省南通环境监测中心 | Deep learning-based multi-scale air quality forecasting method |
| CN111985727A (en) * | 2020-09-03 | 2020-11-24 | 南京信息工程大学 | Weather prediction method and system based on circulation parting model |
| CN112488477A (en) * | 2020-11-23 | 2021-03-12 | 清华珠三角研究院 | Highway emergency management system and method |
| CN115049024A (en) * | 2022-08-15 | 2022-09-13 | 国能日新科技股份有限公司 | Training method and device of wind speed prediction model, electronic equipment and storage medium |
| CN115575920A (en) * | 2022-11-24 | 2023-01-06 | 国家气象中心(中央气象台) | Radar echo dynamic fusion forecasting method, system and equipment based on wavelet transformation |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080097701A1 (en) * | 2006-09-07 | 2008-04-24 | Mcgill University | Short term and long term forecasting systems with enhanced prediction accuracy |
| CN102102626A (en) * | 2011-01-30 | 2011-06-22 | 华北电力大学 | Method for forecasting short-term power in wind power station |
| CN102170130A (en) * | 2011-04-26 | 2011-08-31 | 华北电力大学 | Short-term wind power prediction method |
| CN102184337A (en) * | 2011-06-07 | 2011-09-14 | 中国电力科学研究院 | Dynamic combination analysis method of new energy generating capacity influenced by meteorological information |
| CN102496927A (en) * | 2011-12-16 | 2012-06-13 | 中国电力科学研究院 | Wind power station power projection method based on error statistics modification |
| CN102570453A (en) * | 2012-01-06 | 2012-07-11 | 甘肃省电力公司风电技术中心 | Short-term wind power prediction method and system based on multiple numerical weather prediction sources |
-
2015
- 2015-01-04 CN CN201510004112.9A patent/CN104615855A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080097701A1 (en) * | 2006-09-07 | 2008-04-24 | Mcgill University | Short term and long term forecasting systems with enhanced prediction accuracy |
| CN102102626A (en) * | 2011-01-30 | 2011-06-22 | 华北电力大学 | Method for forecasting short-term power in wind power station |
| CN102170130A (en) * | 2011-04-26 | 2011-08-31 | 华北电力大学 | Short-term wind power prediction method |
| CN102184337A (en) * | 2011-06-07 | 2011-09-14 | 中国电力科学研究院 | Dynamic combination analysis method of new energy generating capacity influenced by meteorological information |
| CN102496927A (en) * | 2011-12-16 | 2012-06-13 | 中国电力科学研究院 | Wind power station power projection method based on error statistics modification |
| CN102570453A (en) * | 2012-01-06 | 2012-07-11 | 甘肃省电力公司风电技术中心 | Short-term wind power prediction method and system based on multiple numerical weather prediction sources |
Non-Patent Citations (3)
| Title |
|---|
| DE GIORGI M G，ET AL.: "《Assessment of the benefits of numerical weather predictions in wind power forecasting based on statistical methods》", 《ENERGY》 * |
| 苏鹏宇: "《考虑风速变化模式的风速预报方法研究》", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 陈欣，等: "《新投产风电场的短期风速预测模型建立》", 《电测与仪表》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105576650B (en) * | 2015-12-29 | 2017-11-21 | 北京国能日新系统控制技术有限公司 | Based on assimilation technique and the wind power forecasting method and system with wind band theory |
| CN105576650A (en) * | 2015-12-29 | 2016-05-11 | 北京国能日新系统控制技术有限公司 | Wind power prediction method and system based on assimilation technology and same wind band theory |
| CN105956252B (en) * | 2016-04-27 | 2019-04-02 | 哈尔滨工业大学 | Based on the multiple dimensioned Forecast model method of ultra-short term wind speed for generating moldeed depth belief network |
| CN105956252A (en) * | 2016-04-27 | 2016-09-21 | 哈尔滨工业大学 | Generative deep belief network-based multi-scale forecast modeling method for ultra-short-term wind speed |
| CN106505631A (en) * | 2016-10-29 | 2017-03-15 | 塞壬智能科技（北京）有限公司 | Intelligent wind power wind power prediction system |
| CN106505631B (en) * | 2016-10-29 | 2019-02-01 | 塞壬智能科技（北京）有限公司 | Intelligent wind power wind power prediction system |
| CN108595658A (en) * | 2018-04-28 | 2018-09-28 | 尚谷科技（天津）有限公司 | A kind of weather knowledge base detection method based on multimedia content |
| CN109214575B (en) * | 2018-09-12 | 2021-08-31 | 河海大学 | Ultrashort-term wind power prediction method based on small-wavelength short-term memory network |
| CN109214575A (en) * | 2018-09-12 | 2019-01-15 | 河海大学 | A kind of super short-period wind power prediction technique based on small wavelength short-term memory network |
| CN111008604A (en) * | 2019-12-09 | 2020-04-14 | 上海眼控科技股份有限公司 | Prediction image acquisition method and device, computer equipment and storage medium |
| CN111339092A (en) * | 2020-02-24 | 2020-06-26 | 江苏省南通环境监测中心 | Deep learning-based multi-scale air quality forecasting method |
| CN111339092B (en) * | 2020-02-24 | 2023-09-08 | 江苏省南通环境监测中心 | Multi-scale air quality forecasting method based on deep learning |
| CN111985727A (en) * | 2020-09-03 | 2020-11-24 | 南京信息工程大学 | Weather prediction method and system based on circulation parting model |
| CN111985727B (en) * | 2020-09-03 | 2023-07-28 | 南京信息工程大学 | Method and system for predicting weather based on loop parting model |
| CN112488477A (en) * | 2020-11-23 | 2021-03-12 | 清华珠三角研究院 | Highway emergency management system and method |
| CN115049024A (en) * | 2022-08-15 | 2022-09-13 | 国能日新科技股份有限公司 | Training method and device of wind speed prediction model, electronic equipment and storage medium |
| CN115575920A (en) * | 2022-11-24 | 2023-01-06 | 国家气象中心(中央气象台) | Radar echo dynamic fusion forecasting method, system and equipment based on wavelet transformation |
| CN115575920B (en) * | 2022-11-24 | 2023-02-28 | 国家气象中心(中央气象台) | Radar echo dynamic fusion forecasting method, system and equipment based on wavelet transformation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104615855A (en) | Day-ahead wind speed multistep prediction method fused with numerical weather prediction | |
| Alessandrini et al. | A novel application of an analog ensemble for short-term wind power forecasting | |
| De Giorgi et al. | Assessment of the benefits of numerical weather predictions in wind power forecasting based on statistical methods | |
| Langodan et al. | A high-resolution assessment of wind and wave energy potentials in the Red Sea | |
| Cadenas et al. | Analysis and forecasting of wind velocity in chetumal, quintana roo, using the single exponential smoothing method | |
| CN103279804B (en) | The Forecasting Methodology of super short-period wind power | |
| Ambach et al. | A new high-dimensional time series approach for wind speed, wind direction and air pressure forecasting | |
| Giannakopoulou et al. | WRF model methodology for offshore wind energy applications | |
| Vincent et al. | Wind fluctuations over the North Sea | |
| Kulkarni et al. | Wind speed prediction using statistical regression and neural network | |
| CN103996079B (en) | Wind power weighting predication method based on conditional probability | |
| Celik | A techno-economic analysis of wind energy in southern Turkey | |
| CN104091077B (en) | A kind of wind energy turbine set photovoltaic plant combines analogy method of exerting oneself | |
| CN103699800A (en) | Ultrashort-period wind speed prediction method based on frequency-domain multi-scale wind speed signal predictability | |
| Kaur et al. | Wind speed forecasting using hybrid wavelet transform-ARMA techniques | |
| Yang et al. | Long-term prediction of significant wave height based on SARIMA model in the South China Sea and adjacent waters | |
| Katikas et al. | A stochastic simulation scheme for the long-term persistence, heavy-tailed and double periodic behavior of observational and reanalysis wind time-series | |
| Lee et al. | Spatial modeling of the highest daily maximum temperature in Korea via max-stable processes | |
| CN105279582B (en) | Super short-period wind power prediction technique based on dynamic correlation feature | |
| Abdulkarim et al. | Statistical analysis of wind speed for electrical power generation in some selected sites in northern Nigeria | |
| Omoyele et al. | Increasing the resolution of solar and wind time series for energy system modeling: A review | |
| Zhao et al. | Wind resource assessment based on numerical simulations and an optimized ensemble system | |
| He et al. | Mapping future offshore wind resources in the South China Sea under climate change by regional climate modeling | |
| Che et al. | An integrated wind-forecast system based on the weather research and forecasting model, Kalman filter, and data assimilation with nacelle-wind observation | |
| Jung et al. | Introducing a new wind speed complementarity model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150513 |
|
| WD01 | Invention patent application deemed withdrawn after publication |