CN103530527A - Wind power probability forecasting method based on numerical weather forecasting ensemble forecasting results - Google Patents

Abstract

The invention provides a wind power probability forecasting method based on numerical weather forecasting ensemble forecasting results. Numerical weather forecasting serves as the foundation, basic input data are provided for short-period wind power forecasting through a numerical weather forecasting ensemble forecasting technology, and a short-period forecasting model is established for each ensemble member to obtain a plurality of groups of forecasting results. For the obtained plurality of groups of forecasting results, and different characteristic forecast errors are classified through an ensemble forecasting configuration characteristic classification method and a forecasting power level classification method to obtain future forecast error bands under certain confidence level. According to the wind power probability forecasting method, under the same confidence level, the error band section is narrower, and for power grids containing large-scale wind power integration, under the condition of the same safety margin, the power grid operation cost can be effectively reduced, and the power grid operation economical property can be improved.

Description

Wind power probability forecasting method based on numerical weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result

Technical field

The present invention relates to wind power prediction field, be specifically related to a kind of wind power probability forecasting method based on numerical weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result.

Background technology

Wind energy is the important component part of regenerative resource, and wind-powered electricity generation is one of renewable energy power generation mode of exploitation the most ripe, on the largest scale at present and commercialized development prospect.But wind-powered electricity generation is different from conventional energy resources, there is very large randomness, intermittence and uncontrollability, large-scale wind-powered electricity generation is connected to the grid, and the planning construction of electrical network, traffic control, analysis and Control, economical operation and the quality of power supply etc. are produced to certain impact.Power Output for Wind Power Field is predicted, it is one of Important Action of reply large-scale wind power access electrical network, wind power prediction can provide technical support for the management and running of electrical network, strengthens the safety and stability of system, is also of value to the formulation of wind energy turbine set operation maintenance plan simultaneously.But wind power has very strong stochastic volatility, the generation rule of wind is difficult to hold, cause wind-powered electricity generation predicated error larger, predict the outcome and be difficult to the formulation of dispatching of power netwoks plan that effective foundation is provided, thereby on wind power deterministic forecast basis, carry out the probabilistic forecasting technical research of wind power, draw the error band interval that predicts the outcome under confidence degree, can be the grid-connected Optimized Operation in large-scale wind power field provides basic technology to support, thereby has important actual application value.

The error band probabilistic forecasting of current wind power mainly adopts definite parameter distribution model, as Gaussian distribution, hyperbolic profile and beta distribution etc., the historical predicated error of matching wind power distributes, the predicated error of unidentified different qualities, resulting probabilistic forecasting error band is identical to all bandwidth that predict the outcome, under identical degree of confidence, width is larger, is unfavorable for the economic load dispatching operation of wind-powered electricity generation.

Summary of the invention

The present invention is directed to the deficiencies in the prior art, a kind of wind power probability forecasting method based on numerical weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result be provided, comprising:

Step 1, each member of logarithm value weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM sets up short-term wind-electricity power forecast model, input respectively the wind power prediction result that obtains member described in each after numerical weather prediction result, obtain the wind power prediction result of the set of weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM;

Step 2, according to the described wind power of each member in described set, predict the outcome and calculate the second order centre distance of described set, according to the differentiation threshold values of the setting of described second order centre distance, identify the error pattern of gathering described in each, the error pattern sequence number of described set is i;

Step 3, divides according to the wind power prediction result of described set the power level of gathering described in each, and the power level sequence number of described set is j;

Step 4, error of calculation type is that i power level is the set { e of relative error of the set of j _ij; Each member's relative error in the set of described relative error

p _mfor revised real power, S _opfor installed capacity;

Step 5, obtains described set { e by the method that core is estimated _ijin the probability density function of each sample;

Step 6, adopts set { e described in the method matching of Nonparametric Fitting Regression _ijprobability density distribution, according to core regression theory, obtain described set { e _ijthe matching regression function m of probability density _ij(e _ij);

Step 7, the matching regression result of the described probability density that described step 6 is obtained returns verification;

Step 8, calculating error pattern is that i power level is the error upper limit and the error lower limit of j underlying menstruation equality when 1-α

with

Step 9, according to the error bound obtaining in described step 8, calculating the error pattern that level of confidence equals 1-α is that i power level is the power prediction result p of j _ijestimation interval Δ p _ij.

In the first preferred embodiment provided by the invention: the wind power prediction result p of the set of the described weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM that described step 1 obtains is

P _mthe wind power that is m set member predicts the outcome, and N represents total number of set member.

In the second preferred embodiment provided by the invention: in described step 2, the second order centre distance of described set

mean value for each ensemble prediction result;

The error pattern of described set is divided into accurately, more accurate and inaccurate 3 classes, and described differentiation threshold values is T ₁and T ₂:

B < T ₁time, the error pattern of set is " accurately ", corresponding error pattern sequence number i=1; T ₁≤ b≤T ₂time, the error pattern of set is " more accurate ", corresponding error pattern sequence number i=2; B > T ₂time, the error pattern of set is " inaccurate ", corresponding error pattern sequence number i=3.

In the 3rd preferred embodiment provided by the invention: in described step 3, the computing method of the power level sequence number j of described set are:

j is positive integer;

η is power level division factor; p _nfor installed capacity, unit is MW;

In the 4th preferred embodiment provided by the invention: in described step 5, the described set { e obtaining by the method that core is estimated _ijin the probability density function of each sample be:

$f_{\mathrm{ij}}\left(e_{\mathrm{ijk}}\right)=\frac{1}{n{\mathrm{\&delta;}}_{\mathrm{ij}}}\underset{M=1}{\overset{n}{\mathrm{\&Sigma;}}}{K}_1\left(\frac{e_{\mathrm{ijk}}-e_{\mathrm{ijM}}}{{\mathrm{\&delta;}}_{\mathrm{ij}}}\right);$

N is set { e _ijin total sample number;

for the kernel function that core is estimated, this kernel function K ₁() gets homogeneous nucleus, δ _ijfor window width; K represents window width δ _ijin the sequence number of sample, f _ij(e _ijk) be window width δ _ijin any one sample e _ijkprobability density function; 1≤M≤n.

In the 5th preferred embodiment provided by the invention: obtain error distributed collection { e according to core regression theory in described step 6 _ijthe matching regression function m of probability density _ij(e _ij) be:

$m_{\mathrm{ij}}\left(e_{\mathrm{ij}}\right)=\left[\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)f_{\mathrm{ij}}\left(e_{\mathrm{ijk}}\right)\right]/\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right);$

Wherein,

for kernel function standard normal core:

$K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)={\left(2\mathrm{\&pi;}\right)}^{-\frac{1}{2}}\exp (-\frac{1}{2}{\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)}^2);$

H _ijfor core returns window width.

In the 6th preferred embodiment provided by the invention: the matching regression result of the probability density that the error that described step 6 is obtained distributes returns verification, the maximum core that is met check results returns window width h _ijvalue:

Checking procedure comprises:

With all error U under i error pattern j power level _ijfor parent, be divided into limited multinomial discrete distribution, establish A ₁..., A _lfor different error event, meet:

wherein, the value of l and h _ijrelevant;

According to described matching regression function m _ij(x) can obtain the probability under different errors:

y _ijk＝m _ij(A _k)·h _ij,k＝1,...,l；

Obtain A _ksampling frequency m under error _ijkmeet

n _ijfor error sample { e _ijcapacity;

Investigate the actual frequency m of sample _ijkto theoretical frequency n _ijy _ijkthe weighted sum of squares of deviation

work as n _ijwhen larger, χ ²the χ that approximate obedience degree of freedom is l-1 ²distribute;

Given level of significance α, if

think that fitting result conforms to actual conditions, have credibility, i.e. verification is passed through, otherwise insincere, and verification is not passed through;

Wherein

value by χ ²distribution upside divides bit table to check in.

In the 7th preferred embodiment provided by the invention: described step 8 comprises:

According to described matching regression function m _ij(e _ij) obtain population distribution function:

$F_{\mathrm{ij}}\left(e\right)={\&Integral;}_{-1}^e{m}_{\mathrm{ij}}\left(e_{\mathrm{ij}}\right)\&CenterDot;h_{\mathrm{ij}}{\mathrm{de}}_{\mathrm{ij}}e\&Element;[-\mathrm{1,1}];$

Wherein, e is power prediction error value, m _ij(e _ij) should meet

Meet $F_{\mathrm{ij}}\left({\mathrm{\&beta;}}_{\mathrm{ij}}^u\right)-F_{\mathrm{ij}}\left({\mathrm{\&beta;}}_{\mathrm{ij}}^l\right)=1-\mathrm{\&alpha;}$ And

value minimum

with

being error pattern is that i power level is that j underlying menstruation equality is in the error upper limit and the error lower limit of 1-α.

In the 8th preferred embodiment provided by the invention: the described estimation interval Δ p in described step 9 _ijfor $\mathrm{\&Delta;}{p}_{\mathrm{ij}}=[p_{\mathrm{ij}}^{\mathrm{lL}},p_{\mathrm{ij}}^{\mathrm{uL}}]:$

$p_{\mathrm{ij}}^{\mathrm{uL}}=\left\{\begin{array}{cc}{p}_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{lL}}\&CenterDot;S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{uL}}\&le;S_{\mathrm{op}}\\ S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{uL}}>S_{\mathrm{op}}\end{array}\right.;$ $p_{\mathrm{ij}}^{\mathrm{lL}}=\left\{\begin{array}{cc}{p}_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{uL}}\&CenterDot;S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{lL}}\&GreaterEqual;0\\ 0& p_{\mathrm{ij}}^{\mathrm{lL}}<0\end{array}\right..$

Compared with the prior art, provided by the invention have a following excellent effect:

1, a kind of wind power probability forecasting method based on numerical weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result provided by the invention, on large component analysis and research actual wind energy turbine set data basis, the main source of finding short-term wind-electricity power predicated error is numerical weather forecast, thereby form, take numerical weather forecast as basis, by numerical weather forecast ensemble forecast technique, be that short-term wind-electricity power is predicted the input data that provide the foundation, for each set member, set up Short-term Forecasting Model, obtaining many groups predicts the outcome, research wind power probabilistic forecasting, under identical level of confidence, error band interval is narrower, for the electrical network containing large-scale wind power access, meeting under identical margin of safety condition, adopt the resulting wind power predicated error of the art of this patent band, can effectively reduce operation of power networks cost, improve the economical of operation of power networks.

2, from wind, belong to meteorological physics research category, having its inherent essential laws attribute sets out, proposition improves the research method of wind power probabilistic forecasting precision by the Classification and Identification of different qualities predicated error, adopt DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM configuration characteristic division methods and predicted power horizontal division method to classify to different qualities predicated error, effectively improved the precision of wind power probabilistic forecasting.

3, adopt distribution-free regression procedure to characterize the probability distribution of the historical predicated error of different qualities, and then draw under confidence degree level following predicated error band, reach by the discriminance analysis of historical predicated error characteristic, hold the object of future anticipation error characteristics.

Accompanying drawing explanation

Be illustrated in figure 1 the process flow diagram of a kind of wind power probability forecasting method based on numerical weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result provided by the invention.

Embodiment

With reference to the accompanying drawings the specific embodiment of the present invention is described in further detail below.

The invention provides a kind of wind power probability forecasting method based on numerical weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result, as shown in Figure 1, as shown in Figure 1, the method comprises its process flow diagram:

Step 1, each member of logarithm value weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM sets up short-term wind-electricity power forecast model, input respectively the wind power prediction result that obtains each member after numerical weather prediction result, obtain the wind power prediction result of the set of weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM.

Step 2, predicts the outcome and calculates the second order centre distance of this set according to the wind power of each member in set, identifies the error pattern of each set according to the differentiation threshold values of the setting of this second order centre distance, the error pattern of each set is divided into three kinds, the error pattern sequence number i=1 gathering, 2,3.

Step 3, divides the power level of each set according to the wind power prediction result p of set, and the power level sequence number of set is j.

Step 4, pair set carry out that error pattern is divided and power level division after, error of calculation type is that i power level is the set { e of relative error of the set of j _ij.Each member's relative error in the set of relative error

p _mfor revised real power, S _opfor installed capacity.

Step 5, obtains { e by the method that core is estimated _ijin the probability density function of each sample.

Step 6, the method matching set { e of employing Nonparametric Fitting Regression _ijprobability density distribution, according to core regression theory, obtain error distributed collection { e _ijthe matching regression function m of probability density _ij(e _ij).

Step 7, the matching regression result of the probability density that the error that step 6 is obtained distributes returns verification.

Step 8, calculating error pattern is that i power level is the error upper limit and the error lower limit of j underlying menstruation equality when 1-α

with

Step 9, it is that i power level is the power prediction result p of j that the error bound obtaining according to step 8 calculates the error pattern that level of confidence equals 1-α _ijestimation interval Δ p _ij.

Further, in step 1, each member of logarithm value weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM sets up short-term wind-electricity power forecast model, inputs respectively the wind power prediction result that obtains each member after numerical weather prediction result, and the wind power prediction result p that obtains the set of weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM is

In formula, p _mthe wind power that is m set member predicts the outcome, and N represents total number of set member.

In step 2, due to cognitive at present finiteness, each member in ensemble prediction all cannot accurately hold the actual change situation of weather, but different initial fields and Parameterization Scheme can be made the reaction that difference is larger to the larger synoptic process of chaotic characteristic, thereby the configuration characteristic that can utilize each set member is held the synoptic process of different qualities, in wind power prediction, show as the predicated error of different qualities, thereby realize the identification division of predicated error type.

According to a large amount of analysis and research, find, the configuration characteristic based on set member can be divided into accurately substantially by predicting the outcome, more accurate and inaccurate 3 classes, and the difference type that predicts the outcome can be judged according to the second order centre distance of DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result.

Each member's the wind power p that predicts the outcome in set _mcalculate the second order centre distance of this set

mean value for each ensemble prediction result.

Suppose that " accurately " and the differentiation threshold values of " more accurate " are T ₁, " more accurate " is T with the differentiation threshold values of " inaccurate " ₂.B < T ₁time, the error pattern of set is " accurately ", corresponding error pattern sequence number i=1; T ₁≤ b≤T ₂time, the error pattern of set is " more accurate ", corresponding error pattern sequence number i=2; B > T ₂time, the error pattern of set is " inaccurate ", corresponding error pattern sequence number i=3.

In step 3, the weather data of numerical weather forecast output is converted into wind energy turbine set predicted power, but wind energy turbine set powertrace has nonlinear characteristic, powertrace plays inhibiting effect at power two ends to the uncertainty of prediction, but the error space is exaggerated, in numerical weather forecast data, exist under error condition, in order to make whole predicated error minimum, predicted data will be compressed into the stage casing part that the error space is relatively little, in case extreme error occurs, make under low wind speed and high wind speed level, the distribution range of predicated error is relatively little, uncertainty is lower, but easily there is extreme error, under medium velocity wind levels, error distribution range is larger, uncertain increasing.Thereby different predicted power levels, the characteristic of predicated error is different.

According to the wind power prediction result p of set, the power level of each set is divided, wherein, the power level sequence number j of set is:

j is positive integer.

η is power level division factor; p _nfor installed capacity, unit is MW;

Being about to power level is divided into following

individual horizontal extent: $[0,\mathrm{\&eta;}{p}_n],(\mathrm{\&eta;}{p}_n,2\mathrm{\&eta;}{p}_n]......(\mathrm{\&eta;}{p}_n\&times;(\frac{1}{\mathrm{\&eta;}}-1),\mathrm{\&eta;}{p}_n\&times;\frac{1}{\mathrm{\&eta;}}],$ Corresponding power level sequence number j is

Power level is got over refinement, more can grasp the inwardness of error, but under the overall stable condition of data, power level refinement is more, corresponding the reducing of data volume in each refinement sample, may cause data sample can not reflect the true distribution of error under respective horizontal, and both produce contradiction.Thereby the degree of refinement of power level, should determine according to concrete data sample and global error situation.

Step 4, pair set carries out, after error pattern division and power level division, calculating relative error collection { e _ij.{ e _ijfor error pattern is that i power level is the set of relative error of the set of j, each member's relative error in the set of relative error

p _mfor revised real power, S _opfor installed capacity.

Step 5, obtains { e by the method that core is estimated _ijin the probability density function of each sample be:

$f_{\mathrm{ij}}\left(e_{\mathrm{ijk}}\right)=\frac{1}{n{\mathrm{\&delta;}}_{\mathrm{ij}}}\underset{M=1}{\overset{n}{\mathrm{\&Sigma;}}}{K}_1\left(\frac{e_{\mathrm{ijk}}-e_{\mathrm{ijM}}}{{\mathrm{\&delta;}}_{\mathrm{ij}}}\right);$

N is set { e _ijin total sample number;

for the kernel function that core is estimated, this kernel function K ₁() can get homogeneous nucleus, δ _ijfor window width; K represents window width δ _ijin the sequence number of sample, f _ij(e _ijk) be window width δ _ijin any one sample e _ijkprobability density function; 1≤M≤n.

Step 6, the method matching set { e of employing Nonparametric Fitting Regression _ijprobability density distribution, according to core regression theory, obtain error distributed collection { e _ijthe matching regression function m of probability density _ij(e _ij):

$m_{\mathrm{ij}}\left(e_{\mathrm{ij}}\right)=\left[\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)f_{\mathrm{ij}}\left(e_{\mathrm{ijk}}\right)\right]/\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right);$

Wherein,

for the kernel function that core returns, h _ijfor core returns window width, this kernel function K () can get standard normal core:

$K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)={\left(2\mathrm{\&pi;}\right)}^{-\frac{1}{2}}\exp (-\frac{1}{2}{\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)}^2).$

Step 7, the matching regression result of the probability density that the error that step 6 is obtained distributes returns verification, and the maximum core that is met check results returns window width h _ijvalue.

Adopt non parametric regression matching to obtain the probability distribution of error, but need carry out verification to fitting of distribution result, could adopt after meeting check results, definite method provided by the invention is to meet h under the prerequisite of test condition _ijget as far as possible higher value.The present invention adopts card side's check addition to carry out verification to fitting result, and checking procedure comprises:

With all error U under i error pattern j power level _ijfor parent, it can be divided into limited multinomial discrete distribution.If A ₁..., A _lfor different error event, meet:

Wherein, the value of l and h _ijrelevant.

According to regression function m _ij(x) can obtain the probability under different errors:

y _ijk＝m _ij(A _k)·h _ij,k＝1,...,l。

{ e _ijbe the error sample under i error pattern j power level, suppose n _ijfor error sample { e _ijcapacity.So can obtain A _ksampling frequency m under error _ijkmeet

$\underset{k=1}{\overset{l}{\mathrm{\&Sigma;}}}{m}_{\mathrm{ijk}}=n_{\mathrm{ij}},k=1,...,l.$

If regression fit result conforms to actual conditions, be positioned at so error A _kunder sampling frequency m _ijkshould be close to n _ijy _ijk.

Investigate the actual frequency m of sample _ijkto theoretical frequency n _ijy _ijkthe weighted sum of squares of deviation

statistic χ ²the size of value has reflected the fitting degree that increment actual frequency distributes theoretical frequency is distributed.Work as n _ijwhen larger, χ ²the χ that approximate obedience degree of freedom is l-1 ²distribute.

Given level of significance α, if think that fitting result conforms to actual conditions, there is credibility, otherwise insincere.Wherein

value can be by χ ²distribution upside divides bit table to check in.

In step 8, when matching distribution results meets after examination requirements, can further set up the error band predicting the outcome, otherwise need to adjust parameter, re-start matching, comprise:

According to matching regression function m _ij(e _ij) obtain population distribution function:

$F_{\mathrm{ij}}\left(e\right)={\&Integral;}_{-1}^e{m}_{\mathrm{ij}}\left(e_{\mathrm{ij}}\right)\&CenterDot;h_{\mathrm{ij}}{\mathrm{de}}_{\mathrm{ij}}e\&Element;[-\mathrm{1,1}].$

Wherein, e is power prediction error value, m _ij(e _ij) should meet

Meet

's

with

have or not array, but spacing minimum only has one group, referred to here as bandwidth minimum principle.Meet $F_{\mathrm{ij}}\left({\mathrm{\&beta;}}_{\mathrm{ij}}^u\right)-F_{\mathrm{ij}}\left({\mathrm{\&beta;}}_{\mathrm{ij}}^l\right)=1-\mathrm{\&alpha;}$ And

value hour,

with

be that error pattern is that i power level is that j underlying menstruation equality is in the error upper limit and the error lower limit of 1-α.

Step 9, it is that i power level is the power prediction result p of j that the error bound obtaining according to step 8 calculates the error pattern that level of confidence equals 1-α _ijestimation interval Δ p _ij:

Concrete, after establishing error upper and lower bound, error burst need to be changed into predicted power interval, according to $e=\frac{P-P_M}{S_{\mathrm{op}}}$ Can obtain: $\left\{\begin{array}{c}{p}_{\mathrm{ij}}^{\mathrm{uL}}=p_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{lL}}\&CenterDot;S_{\mathrm{op}}\\ p_{\mathrm{ij}}^{\mathrm{lL}}=p_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{uL}}\&CenterDot;S_{\mathrm{op}}\end{array}\right..$

In formula, p _ijfor the predicted power under i error pattern j power level.Power upper limit

may exceed installed capacity S _oprestriction, but in fact power upper limit can not surpass installed capacity, thereby

$p_{\mathrm{ij}}^{\mathrm{uL}}=\left\{\begin{array}{cc}{p}_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{lL}}\&CenterDot;S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{uL}}\&le;S_{\mathrm{op}}\\ S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{uL}}>S_{\mathrm{op}}\end{array}\right..$

Equally, power lower limit

may become negative, so

$p_{\mathrm{ij}}^{\mathrm{lL}}=\left\{\begin{array}{cc}{p}_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{uL}}\&CenterDot;S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{lL}}\&GreaterEqual;0\\ 0& p_{\mathrm{ij}}^{\mathrm{lL}}<0\end{array}\right..$

Through actual wind energy turbine set test, show, the method can effectively draw the error band that wind power predicts the outcome, and by finding with current main probability forecasting method contrast, under identical level of confidence, institute's extracting method can obtain the more predicated error band of high accurancy and precision herein.

Finally should be noted that: above embodiment is only in order to illustrate that technical scheme of the present invention is not intended to limit, although the present invention is had been described in detail with reference to above-described embodiment, those of ordinary skill in the field are to be understood that: still can modify or be equal to replacement the specific embodiment of the present invention, and do not depart from any modification of spirit and scope of the invention or be equal to replacement, it all should be encompassed in the middle of claim scope of the present invention.

Claims (9)

1. the wind power probability forecasting method based on numerical weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM result, is characterized in that, described method comprises:

Step 1, each member of logarithm value weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM sets up short-term wind-electricity power forecast model, input respectively the wind power prediction result that obtains member described in each after numerical weather prediction result, obtain the wind power prediction result of the set of weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM;

Step 2, according to the described wind power of each member in described set, predict the outcome and calculate the second order centre distance of described set, according to the differentiation threshold values of the setting of described second order centre distance, identify the error pattern of gathering described in each, the error pattern sequence number of described set is i;

Step 3, divides according to the wind power prediction result of described set the power level of gathering described in each, and the power level sequence number of described set is j;

Step 4, error of calculation type is that i power level is the set { e of relative error of the set of j _ij; Each member's relative error in the set of described relative error

p _mfor revised real power, S _opfor installed capacity;

Step 5, obtains described set { e by the method that core is estimated _ijin the probability density function of each sample;

Step 6, adopts set { e described in the method matching of Nonparametric Fitting Regression _ijprobability density distribution, according to core regression theory, obtain described set { e _ijthe matching regression function m of probability density _ij(e _ij);

Step 7, the matching regression result of the described probability density that described step 6 is obtained returns verification;

Step 8, calculating error pattern is that i power level is the error upper limit and the error lower limit of j underlying menstruation equality when 1-α

with

Step 9, according to the error bound obtaining in described step 8, calculating the error pattern that level of confidence equals 1-α is that i power level is the power prediction result p of j _ijestimation interval Δ p _ij.

2. the method for claim 1, is characterized in that, the wind power prediction result p of the set of the described weather forecast DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM that described step 1 obtains is

P _mthe wind power that is m set member predicts the outcome, and N represents total number of set member.

3. method as claimed in claim 2, is characterized in that, in described step 2, and the second order centre distance of described set mean value for each ensemble prediction result;

The error pattern of described set is divided into accurately, more accurate and inaccurate 3 classes, and described differentiation threshold values is T ₁and T ₂:

B < T ₁time, the error pattern of set is " accurately ", corresponding error pattern sequence number i=1; T ₁≤ b≤T ₂time, the error pattern of set is " more accurate ", corresponding error pattern sequence number i=2; B > T ₂time, the error pattern of set is " inaccurate ", corresponding error pattern sequence number i=3.

4. method as claimed in claim 2, is characterized in that, in described step 3, the computing method of the power level sequence number j of described set are:

j is positive integer;

η is power level division factor; p _nfor installed capacity, unit is MW;

5. the method for claim 1, is characterized in that, in described step 5, and the described set { e obtaining by the method that core is estimated _ijin the probability density function of each sample be:

$f_{\mathrm{ij}}\left(e_{\mathrm{ijk}}\right)=\frac{1}{n{\mathrm{\&delta;}}_{\mathrm{ij}}}\underset{M=1}{\overset{n}{\mathrm{\&Sigma;}}}{K}_1\left(\frac{e_{\mathrm{ijk}}-e_{\mathrm{ijM}}}{{\mathrm{\&delta;}}_{\mathrm{ij}}}\right);$

N is set { e _ijin total sample number;

for the kernel function that core is estimated, described kernel function K ₁() gets homogeneous nucleus, δ _ijfor window width; K represents window width δ _ijin the sequence number of sample, f _ij(e _ijk) be window width δ _ijin any one sample e _ijkprobability density function; 1≤M≤n.

6. method as claimed in claim 5, is characterized in that, in described step 6, according to core regression theory, obtains error distributed collection { e _ijthe matching regression function m of probability density _ij(e _ij) be:

$m_{\mathrm{ij}}\left(e_{\mathrm{ij}}\right)=\left[\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)f_{\mathrm{ij}}\left(e_{\mathrm{ijk}}\right)\right]/\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right);$

Wherein, for kernel function standard normal core:

$K\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)={\left(2\mathrm{\&pi;}\right)}^{-\frac{1}{2}}\exp (-\frac{1}{2}{\left(\frac{e_{\mathrm{ij}}-e_{\mathrm{ijk}}}{h_{\mathrm{ij}}}\right)}^2);$

H _ijfor core returns window width.

7. method as claimed in claim 6, is characterized in that, the matching regression result of the probability density that the error that described step 6 is obtained distributes returns verification, and the maximum core that is met check results returns window width h _ijvalue:

Checking procedure comprises:

With all error U under i error pattern j power level _ijfor parent, be divided into limited multinomial discrete distribution, establish A ₁..., A _lfor different error event, meet:

wherein, the value of l and h _ijrelevant;

According to described matching regression function m _ij(x) can obtain the probability under different errors:

y _ijk＝m _ij(A _k)·h _ij,k＝1,...,l；

Obtain A _ksampling frequency m under error _ijkmeet n _ijfor error sample { e _ijcapacity;

Investigate the actual frequency m of sample _ijkto theoretical frequency n _ijy _ijkthe weighted sum of squares of deviation work as n _ijwhen larger, χ ²the χ that approximate obedience degree of freedom is l-1 ²distribute;

Given level of significance α, if

think that fitting result conforms to actual conditions, have credibility, i.e. verification is passed through, otherwise insincere, and verification is not passed through;

Wherein

value by χ ²distribution upside divides bit table to check in.

8. method as claimed in claim 6, is characterized in that, described step 8 comprises:

According to described matching regression function m _ij(e _ij) obtain population distribution function:

$F_{\mathrm{ij}}\left(e\right)={\&Integral;}_{-1}^e{m}_{\mathrm{ij}}\left(e_{\mathrm{ij}}\right)\&CenterDot;h_{\mathrm{ij}}{\mathrm{de}}_{\mathrm{ij}}e\&Element;[-\mathrm{1,1}];$

Wherein, e is power prediction error value, m _ij(e _ij) should meet

Meet $F_{\mathrm{ij}}\left({\mathrm{\&beta;}}_{\mathrm{ij}}^u\right)-F_{\mathrm{ij}}\left({\mathrm{\&beta;}}_{\mathrm{ij}}^l\right)=1-\mathrm{\&alpha;}$ And

value minimum

with

being error pattern is that i power level is that j underlying menstruation equality is in the error upper limit and the error lower limit of 1-α.

9. method as claimed in claim 6, is characterized in that, the described estimation interval Δ p in described step 9 _ijfor $\mathrm{\&Delta;}{p}_{\mathrm{ij}}=[p_{\mathrm{ij}}^{\mathrm{lL}},p_{\mathrm{ij}}^{\mathrm{uL}}]:$

$p_{\mathrm{ij}}^{\mathrm{uL}}=\left\{\begin{array}{cc}{p}_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{lL}}\&CenterDot;S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{uL}}\&le;S_{\mathrm{op}}\\ S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{uL}}>S_{\mathrm{op}}\end{array}\right.;$ $p_{\mathrm{ij}}^{\mathrm{lL}}=\left\{\begin{array}{cc}{p}_{\mathrm{ij}}-{\mathrm{\&beta;}}_{\mathrm{ij}}^{\mathrm{uL}}\&CenterDot;S_{\mathrm{op}}& p_{\mathrm{ij}}^{\mathrm{lL}}\&GreaterEqual;0\\ 0& p_{\mathrm{ij}}^{\mathrm{lL}}<0\end{array}\right..$

Images