CN109886452A - A kind of ultrashort-term wind power probability forecasting method and system based on experience dynamic modeling - Google Patents
A kind of ultrashort-term wind power probability forecasting method and system based on experience dynamic modeling Download PDFInfo
- Publication number
- CN109886452A CN109886452A CN201910009512.7A CN201910009512A CN109886452A CN 109886452 A CN109886452 A CN 109886452A CN 201910009512 A CN201910009512 A CN 201910009512A CN 109886452 A CN109886452 A CN 109886452A
- Authority
- CN
- China
- Prior art keywords
- wind power
- ultrashort
- dynamic modeling
- term wind
- method based
- 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.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a kind of ultrashort-term wind power probability forecasting methods and system based on experience dynamic modeling, wherein this method comprises: treating premeasuring time series carries out standard normal processing, and non-linear degree of polymerization calculating is carried out to standard normalization treated data, to investigate the nonlinear degree of given dynamical system；Using particle swarm optimization algorithm, optimal embedding dimension E and delay time T are calculated；Further, it treats premeasuring time series and carries out phase space reconfiguration；Empirical dynamic model is constructed, given dynamical system is predicted using simplex sciagraphy in phase space reconstruction, obtains the prediction result of amount to be predicted.Prediction result is shown, can be realized to wind-power electricity generation dynamic process using the ultrashort-term wind power probability forecasting method based on experience dynamic modeling completely according to the objective description of data, has been obviously improved the validity of probabilistic forecasting.
Description
Technical field
The invention belongs to wind power prediction technical field during generation of electricity by new energy, more particularly to one kind are dynamic based on experience
The ultrashort-term wind power probability forecasting method and system of state modeling.
Background technique
Wind-power electricity generation is the most mature renewable energy power generation form of current development, accesses modern power network on a large scale,
It is being that the energy-saving and emission-reduction cause of the whole society makes significant contribution.Representative of the wind-powered electricity generation as generation of electricity by new energy relies on China unique
Geographical features and policy superiority, possess powerful growth momentum and vast market prospect.At the same time, due to wind power plant
The extensive access of the fluctuation and intermittence of output power, wind-powered electricity generation brings huge challenge to the traffic control of electric system.It is right
It is to alleviate electric system peak regulation, frequency modulation pressure that wind power plant, which carries out power prediction, improves the effective means that wind-powered electricity generation receives ability, even more
It instructs wind power plant to formulate maintenance plan, improves wind energy utilization, promote the effective way of economic benefit.
According to the difference of predetermined period, wind power prediction can be divided into medium-term and long-term, short-term and ultra-short term prediction.Wind power
Long-term forecast using year as timeliness, be chiefly used in the planning of wind power plant and formulate annual generation schedule；The mid-term of wind power is pre-
Survey the maintenance plan for being chiefly used in formulating wind power plant using week or the moon as timeliness.Requirement of the medium- and long-term forecasting to precision of prediction be not tight
Lattice, but need to accumulate prolonged operation data.The short-term forecast of wind power is usually with 1~3 day for predetermined period, in order to
Reduce and even avoid abandonment, short-term forecast to the more demanding of precision, be usually used in optimizing the daily trading plannings of normal power supplies with it is cold
The activities such as stand-by heat, and adjustment maintenance plan.Ultra-short term prediction then refers to pre- to 0~4 hour following progress wind power
It surveys, ultra-short term prediction helps to optimize spinning reserve capacity and electric system frequency modulation and voltage modulation, is easy to implement on-line optimization unit
Combination and economic load dispatching.
According to the difference of prediction result form, wind power forecasting method can be divided into monodrome (or certainty) prediction and
Probability (or uncertain) two classes of prediction.Wind power prediction technology used at present is mostly monodrome prediction technique, is only predicted
The following Power Output for Wind Power Field an of conditional expectation is obtained, is a kind of deterministic forecast.In order to improve monodrome prediction technique
Accuracy, domestic and foreign scholars have been carried out numerous studies, and still, deterministic forecast method is difficult to have on precision of prediction larger prominent
It is broken, because the total data for wanting to obtain future event is unpractical.Therefore, any prediction technique has it intrinsic not really
It is qualitative and can not be restrictive, it is impossible to all information of future event are obtained, realize accurate prediction, it is especially capricious big
Gas behavior.Compared with deterministic prediction, the uncertain prediction technique for providing the probabilistic information of future event is more accurate than realizing
Prediction is advantageously.Probabilistic forecasting uncertain prediction in other words has significant progress with the development of theoretical prediction.
From the perspective of physics, wind power plant can be considered as with the artificial physical system for determining dynamic characteristic.Although
In this way, Power Output for Wind Power Field but has stronger uncertainty, basic reason is the non-of wind power plant dynamical system itself
Linearly, the uncertainty of complexity and boundary conditions, these characteristics embody a concentrated expression of following aspect.Firstly, wind power plant
Structure is complicated, and there are apparent nonlinear characteristics.Under certain meteorological condition, the landform of the output power of wind power plant and its inside
The many factors such as looks, blower layout, blower output characteristics are related, between these influence factors, influence factor and wind power plant it is defeated
There is apparent non-linear correlation relationships (to deposit between wind speed and blower output power for example, generally believing again between power out
In high order, noncontinuous map relationship).Thus wind power plant dynamical system complexity with higher is determined, and along with non-thread
Property feature, output power is more sensitive to boundary meteorological condition, difficult due to existing even for seeming identical meteorological condition
In terms of and nuance, there is also apparent fluctuations for output power.Secondly as complicated, the non-linear spy of Atmosphere System
Sign, wind-powered electricity generation field border meteorological condition are difficult to accurately obtain.It can be appreciated that Atmosphere System itself is that have the presence of strong nonlinearity mixed
The complication system of ignorant phenomenon, the prediction to its future developing trend, there are stronger uncertainties.However, wind power plant dynamical system
The estimation of state variable variation track is needed according to boundary condition provided by Atmosphere System in system, to the estimation of this boundary condition
Inaccuracy, so that the complication and non-linearity feature of Atmosphere System retains and amplifies in wind power plant dynamical system.Thus it is apparent from,
For uncertain so high wind power plant dynamical system, it is desirable to which it is extremely difficult for going description using the fixed model of equation
, fixed equation limits the excavation of effective information, thus when meteorological condition changes, it is difficult to provide accurately pre-
It surveys.
However, wind power plant in the operational process of its construction, trial operation and the wind power plant that put into operation, has accumulated largely
Meteorological and operation/maintenance data has contained the sufficient behavioral characteristics of the system in the behavioral data of wind power plant.It is thereby achieved that
The behavioral characteristics assumed without prior model based on measured data excavate, the understanding for this nonlinear dynamic system of wind power plant
It is most important with prediction.Therefore, this patent using based on empirical dynamic model without prediction equation, it is intended to abundant mining data
Feature is hidden, its development track is described, to carry out ultra-short term probabilistic forecasting to wind power.
Summary of the invention
The present invention provides a kind of ultrashort-term wind power probability forecasting method and its system based on experience dynamic modeling,
High for wind power plant dynamical system complexity its object is to solve, model bias is unfavorable for the excavation of effective information in data,
It is difficult to the technical issues of providing high-precision wind-powered electricity generation prediction result in practice.
In order to solve the above-mentioned technical problem, the present invention proposes that a kind of ultrashort-term wind power based on experience dynamic modeling is general
Rate prediction technique, comprising:
Step (1): the time series for treating premeasuring carries out standard normal processing and carries out to the data put in order non-
The calculating of linear polymerization degree, to investigate the nonlinear degree of system；
Step (2): using particle swarm optimization algorithm, and optimizing obtains Embedded dimensions E and delay time T；
Step (3): according to Embedded dimensions E and delay time T required by step (1), (2), the time sequence of premeasuring is treated
Column carry out phase space reconfiguration；
Step (4): building empirical dynamic model is predicted, the prediction for obtaining wind power is general in phase space reconstruction
Rate distribution；
Step (5): according to selected confidence level, the bound of wind-powered electricity generation prediction power output, i.e. wind-powered electricity generation prediction power output section are obtained
As a result.
Step (1) the Plays normal state is when being converted to amount time series to be predicted to obey standardized normal distribution
Between sequence, even if it is desired for 0, variance 1:
Vt'=(Vt-μ(Vt))/σ(Vt) (1)
In formula, VtFor the former time series of amount to be predicted, μ (Vt) be the time series expectation, σ (Vt) it is the time sequence
The standard deviation of column, Vt' it is time series after standard normal.
The calculating of the non-linear degree of polymerization is the method using S mapping in the step (1).S mapping calculation is lag
Locally linear embedding between coordinate vector and target variable, it include a Dynamic gene θ, the factor be used to control it is each to
The weight contacted between amount: S is mapped as linear autoregressive models when θ=0；And θ > 0 is then assigned when calculating locally linear embedding
The more weights of adjacent states amount, to show non-linear.
Embedded dimensions E and delay time T are calculated in the step (2) using particle swarm optimization algorithm, in population
In algorithm, the solution of each optimization problem is one " particle " in search space.Each particle has an initialization speed
And position, an adaptive value determined by fitness function.Each particle is endowed memory function, can remember to search most
Best placement, furthermore the speed of each particle determines the direction and distance that they are searched for, so that particle can be in optimal solution space
Search.In iteration searching process each time, particle updated by comparing fitness value and two extreme values oneself speed and
Position: optimal solution (the individual extreme value p that particle itself is foundbest) and the optimal solution (global extremum that finds at present of entire population
gbest), i.e.,
xi(t+1)=xi(t)+vi(t+1) (3)
Wherein, t represents the t times iteration, viRepresent the speed of i-th of particle, xiThe position of i-th of particle is represented, ω is used
Property weight；c1And c2For perception factor, R1And R2It is two random numbers in [0,1] range.In the step, population is calculated
The optimization aim of method is to make to cover bandwidth index (CWC) minimum, this is also the method for evaluating particle populations fitness.
The phase space reconfiguration of the step (3), which refers to, carries out weight to former nonlinear dynamic system with the timing observation of one-dimensional
Structure, to describe its Evolution and developing state.According to Ta Kensi theorem and Whitney embedding theorems, for chaos system
One-dimensional observation sequence, as long as Embedded dimensions E meets E >=2M+1 (M is motive power system dimension), so that it may obtain one and original
The reconfiguration system of system differential homeomorphism, i.e., former dynamical system can be reconstructed by the timing observation of one-dimensional observed quantity.Therefore, only
Choose suitable Embedded dimensions E and delay time T, so that it may work as motive power system reconfiguration to a higher-dimension phase space
In, the analysis to original system is realized in reconstruction attractor.
The prediction process of the step (4) uses simplex sciagraphy.Simplex sciagraphy is that time delay is embedding
Enter into a single time series to generate attractor reconstruction, predicted in phase space reconstruction, principle is sketched such as
Under: simplex sciagraphy predicts what current quantity of state was likely to occur in next step by calculating the motion profile of quantity of state consecutive points
Motion profile.A given phase space reconstruction and a quantity of state Xs, X is found firstsB consecutive points (the usually setting b of surrounding
=E+1, E are Embedded dimensions), these consecutive points are denoted as quantity of state Xn(s,i), wherein n (s, i) indicates distance XsI-th it is close when
Sequence observation, i.e. Xn(s,1)It is distance XsNearest point, Xn(s,2)It is distance XsSecond close point, and so on；Then it observes and remembers
The variation track of these consecutive points is recorded, the track of each consecutive points can regard quantity of state X assThe following possible motion profile,
Therefore, the position X after h time step of each consecutive pointsn(s,i)+hCertain probability is XsPosition after h time step
Xs+h, wherein each Xn(s,i)+hProbability P (i) be:
Wherein, d (Xs,Xn(s,i)) it is quantity of state XsWith quantity of state Xn(s,i)Between Euclidean distance, N be consecutive points number N
=b, b=E+1.Power probability distribution table according to each future value and its probability of appearance, after available h step-length；Into
One step, when consecutive points quantity is sufficiently large, so that it may which approximation obtains the power probability distribution after h time step.
Step (5) basis confidence level selected in advance is (such as: 90%), from step (4) obtained power probability
The upper and lower section that prediction power is obtained in distribution, selects the available upper and lower section of different power of different confidence levels,
To obtain the confidence belt of different in width.
Beneficial technical effect
1, as a kind of prediction technique based on data mining, the present invention is using the historical time sequence of amount to be predicted as instruction
Practice data, the wide area measurement information in wind power plant can be made full use of；Meanwhile model considers adjacent node in learning process
Influence effectiveness to node to be predicted, therefore can reflect the space time correlation characteristic of electric system, to more accurately describe state
The changing rule and operation situation of amount, reduce prediction error, obtain more structurally sound prediction result.
2, as a kind of dynamic modelling method, the present invention is not limited by preset parameter equation, can be according to quantity of state
Variation tendency is constantly adjusted model, so that model has very strong adaptability.Therefore, under steady state conditions, mould
Type can be automatically adjusted to optimum state according to the variation of data, to ensure that high-precision prediction result.
3, as a kind of probability forecasting method, the present invention is in the base for realizing the single desired value prediction of wind power plant future generated energy
On plinth, also reliable prediction can be carried out for the waving interval of prediction error, is dispatching of power netwoks and power train under specified confidence level
The reliability service of system provides more comprehensively predictive information.
Detailed description of the invention
The accompanying drawings constituting a part of this application is provided for further understanding of the present application, and combines this
The illustrative examples of application are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is the ultrashort-term wind power probability forecasting method block diagram of the invention based on experience dynamic modeling；
Fig. 2 is phase space reconfiguration schematic diagram of the present invention；
Fig. 3 is the ultrashort-term wind power probability forecasting method procedure chart of the invention based on experience dynamic modeling；
Fig. 4 is simplex sciagraphy prediction principle figure of the present invention；
Fig. 5 is case verification wind field interval prediction result figure in the present invention；
Fig. 6 is the probabilistic forecasting result figure in present example verifying；
Fig. 7 is the structural representation of the ultrashort-term wind power probabilistic forecasting system of the invention based on experience dynamic modeling
Figure.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
The embodiment of the present invention has carried out the super of 15 minutes futures to the wind power of Shandong Area in Yantai Region Peng Lai wind power plant
Short term probability prediction, the invention will be further described with embodiment with reference to the accompanying drawing.
As shown in Figure 1, the wind power ultra-short term probability forecasting method based on experience dynamic modeling, mainly includes as follows
Step:
Step (1): data preparation and processing --- treat the time sequence of premeasuring, such as power, voltage magnitude, phase angle etc.
Column carry out standard normal processing, and the calculating to standard normalization treated data the carry out non-linear degree of polymerization, to investigate
The nonlinear degree of given dynamical system.
Firstly, the time series of amount to be predicted is carried out standard normal processing according to step (1).Wherein, standard normal
Change is that amount time series to be predicted is converted to the time series for obeying standardized normal distribution, even if it is desired for 0, variance 1:
Vt'=(Vt-μ(Vt))/σ(Vt) (5)
In formula, VtFor the former time series of amount to be predicted, μ (Vt) be the time series expectation, σ (Vt) it is the time sequence
The standard deviation of column, Vt' it is time series after standard normal.
Then, non-linear polymerization is carried out to the amount time series to be predicted after standard normal to calculate.
The calculating of the non-linear degree of polymerization is the method using S mapping.S mapping calculation is the coordinate vector lagged and target
Locally linear embedding between variable, including a Dynamic gene θ, the factor are used to control the power contacted between each vector
Weight: S is mapped as linear autoregressive models when θ=0；And θ > 0 then imparts adjacent states amount more when calculating locally linear embedding
More weight, to show non-linear.θ > 0 is obtained by calculation, illustrates that power time series have non-linear behavior.
Step (2): using particle swarm algorithm, calculates optimal embedding dimension E and delay time T.
The behavior that the thought of particle swarm algorithm is looked for food derived from the research to flock of birds predation, simulation bird cluster flight, bird
Between so that group is optimal purpose by the cooperation of collective, be a kind of optimization method for being based on " group intelligence ".In population
In algorithm, the solution of each optimization problem is one " particle " in search space.Each particle has an initialization speed
And position, an adaptive value determined by fitness function.Each particle is endowed memory function, can remember to search most
Best placement simultaneously shares this optimum position with group, and furthermore the speed of each particle determines the direction and distance that they are searched for, with
Just particle can be searched in optimal solution space.In iteration searching process each time, particle is by comparing fitness value and two
A extreme value updates oneself speed and position: optimal solution (the individual extreme value p that particle itself is foundbest) and entire population mesh
Before optimal solution (the global extremum g that findsbest), i.e.,
xi(t+1)=xi(t)+vi(t+1) (7)
Wherein, t represents the t times iteration, viRepresent the speed of i-th of particle, xiThe position of i-th of particle is represented, ω is used
Property weight；c1And c2For perception factor, R1And R2It is two random numbers in [0,1] range.
In the step, the optimization aim of particle swarm algorithm is to make to cover bandwidth index (CWC) minimum, this is also to be used for
The method for evaluating particle populations fitness.Corresponding Embedded dimensions E and delay time T are required when CWC minimum.
Covering bandwidth index is a common composite target for measuring forecast interval quality, is worth smaller, it is meant that pre-
It is better to survey section.Covering bandwidth index consists of two parts: section covering power (PICP) and interval width (PINAW).
Section covering power refers to that gained forecast interval includes the ability of true value, and value is bigger, it is meant that forecast interval
Reliability it is higher, its calculation formula is:
Wherein, c (i) is the indicator function of coverage, expression formula are as follows:
Wherein, yiFor i-th of observation, [Li,Ui] be i-th of forecast interval up/down circle, M be forecast interval number.
Interval width then means the acutance in section, the forecast interval under identical confidence level, the smaller person of interval width,
The information of reflection is abundanter, and acutance is also better.The calculation formula of interval width are as follows:
Wherein M is forecast interval number.
Cover the calculation formula of bandwidth index are as follows:
CWC=PINAW (1+ γ (PICP) e-η(PICP-μ)) (11)
Wherein, μ is preset confidence level, and taking 0.9, η in present example is penalty factor, of the invention real
It is jump function, expression formula that 10, γ (PICP) is taken in example are as follows:
By above-mentioned calculating, the x obtained by formula (7)i(t+1) be particle a position, the position be it is two-dimensional, i.e.,
It is made of two parameters of E and τ；According to particle xi(t+1) E and τ when carry out phase space reconfiguration, then carry out in reconstruction attractor
Prediction, the result and observation y of predictioniIt is compared, and is judged with CWC；The phase space that different E and τ is constituted is not
Together, therefore prediction result and CWC score are also different, and the E and τ when selecting prediction effect optimal (that is: CWC minimum) are as optimal
Solution.
Step (3): it according to optimal embedding dimension E and delay time T, treats premeasuring time series and carries out phase space weight
Structure.
Refer in the phase space reconfiguration of step (3) and weight is carried out to former nonlinear dynamic system with the timing observation of one-dimensional
Structure, to describe its Evolution and developing state.According to Ta Kensi theorem, as long as optimal embedding dimension E meets E >=2M+1 (M
For motive power system dimension) when, given dynamical system can be reconstructed by the timing observation of one-dimensional observed quantity.Such as Fig. 2 institute
Show, Fig. 2 (a) is geometric locus of the dynamical system in three-dimensional state space, and curve has been integrally formed this system in state space
Interior manifold, according to Ta Kensi theorem, which can be reconstructed by the timing observation of one-dimensional observed quantity in system.Fig. 2 (b) is
Former dynamical system is by the manifold after the timing observation reconstruct of one-dimensional variable y, wherein τ is hour described in Ta Kensi theorem
Between be spaced.Fig. 2 (c) gives the dynamical system by the manifold after the timing observation reconstruct of observed quantity y, z.Obtained by Fig. 2, only
Choose suitable optimal embedding dimension E and delay time T, so that it may by motive power system reconfiguration to a higher-dimension phase space
In the middle, the analysis to original system is realized in reconstruction attractor.
According to Embedded dimensions E and delay time T, phase space reconfiguration is carried out to Yantai Peng Lai wind power time series,
Model expression are as follows:
Xt=< Vt> (13)
Wherein XtFor by timing observation { xtConstruction E dimensional vector, Xt=< xt,xt-τ,xt-2τ,...,xt-(E-1)τ>, t is
Measure time, VtFor Yantai Peng Lai wind power observation sequence.
Step (4): constructing empirical dynamic model, predict in phase space reconstruction Yantai Peng Lai wind power,
Obtain prediction result.
During prediction in phase space reconstruction given dynamical system, calculated using simplex sciagraphy
The motion profile of quantity of state consecutive points, and the probability occurred to every track is estimated, and then predicts given dynamical system
The motion profile of state.
Simplex sciagraphy is to be embedded into time delay in one single time series to generate attractor reconstruction, from
And it is predicted.Specifically, wind-powered electricity generation future time instance is predicted by the probability of occurrence of the motion profile of calculating quantity of state consecutive points
Power probability distribution.
A given phase space reconstruction and a quantity of state Xs, X is found firstsB consecutive points (the usually setting b=of surrounding
E+1, E are Embedded dimensions), these consecutive points are denoted as vector Xn(s,i), wherein n (s, i) indicates distance XsI-th close timing is seen
Measured value, i.e. Xn(s,1)It is distance XsNearest point, Xn(s,2)It is distance XsSecond close point, and so on；Then this is observed and recorded
The variation track of a little consecutive points, the track of each consecutive points can regard quantity of state X assThe following possible motion profile, therefore,
Position X after h time step of each consecutive pointsn(s,i)+hCertain probability is XsPosition X after h time steps+h,
Wherein, each Xn(s,i)+hProbability P (i) be:
Wherein, d (Xs,Xn(s,i)) it is quantity of state XsWith quantity of state Xn(s,i)Between Euclidean distance；N is the number of consecutive points,
N=b, b=E+1.Power probability distribution table according to each future value and its probability of appearance, after available h step-length；Into
One step, when consecutive points quantity is sufficiently large, so that it may which approximation obtains the power probability distribution after h time step.
Further, according to selected in advance confidence level (such as: 90%), the obtained power probability from above-mentioned steps
The upper and lower section that prediction power is obtained in distribution, selects the available upper and lower section of different power of different confidence levels,
To obtain the confidence belt of different in width.
Fig. 3 illustrates the prediction for carrying out ultrashort-term wind power probabilistic forecasting to Yantai Peng Lai wind power plant using the method
Process: firstly, treating premeasuring time series carries out phase space reconfiguration, and then simplex sciagraphy is used in phase space reconstruction
Given dynamical system is predicted, predicts wind-powered electricity generation not by the probability of occurrence of the motion profile of calculating quantity of state consecutive points
Carry out the distribution of moment power probability.
Fig. 4 then further illustrates the prediction process of simplex sciagraphy.
Yantai Peng Lai wind power time series are chosen as sample, sample length is, time interval is 5 points at 8000 points
Clock.Carry out 5 minutes following, 15 minutes futures, and prediction in following 30 minutes respectively to sample, and using the side of cross validation
Formula is tested.In this example, a point seasonal forecasting is carried out to sample, wherein -2 months December was winter, and the 3-5 month is spring, the 6-8 month
For summer, the 9-11 month is autumn.
Cross validation is meant that in given sample, the major part for taking out sample is modeled, and fraction is used to examine
The accuracy of model built is tested, and records prediction error.The embodiment of the present invention uses four times of cross validation schemes to model: will count
According to being equally divided into four sections, 2000 points every section, successively chooses three sections and be used for training pattern, remaining one section is used to predict, in this process
In, model has carried out primary prediction to every section of sample standard deviation.
Precision evaluation index: the present invention will cover bandwidth index as precision evaluation index, and calculation formula is shown in (11).
It is as shown in table 1 using optimal embedding dimension E obtained by particle swarm algorithm and delay time T, E=7, τ=5 can be obtained.
1 population calculated result of table
It is as shown in table 2 using the prediction result of the method for the present invention:
2 invention example prediction result of table is shown
For covering power PICP, value is bigger, and the covering power for representing forecast interval is stronger；For interval width
For PINAW, value is smaller, and the width for representing forecast interval is narrower, shows that the practicability of forecast interval is better；Synthesis is covered
For lid bandwidth index CWC, value is smaller, shows that the comprehensive performance of forecast interval is better.As can be seen from Table 2, the value of CWC with
The extension of prediction scale and increase, this is mainly due to the extension with prediction scale, the uncertain enhancing of wind-powered electricity generation, prediction
Section broadens, so that PINAW score increases, so as to cause the increase of CWC value.
The method of the present invention is as shown in table 3 compared with the prediction result of other prediction techniques:
The comparison of 3 prediction result of table
As can be seen from Table 3, the CWC value of the resulting forecast interval of this method is minimum, shows to carry out using the method for the present invention
Prediction can obtain more structurally sound prediction result, therefore, it is advantageous carry out data mining to effective metrical information in wind power plant
In the precision for improving load prediction.The embodiment of the present invention only predicts the active power of wind power plant, in fact, side of the present invention
Method can be widely applied to the prediction of other electrical quantity (voltage magnitude, phase angle) in power grid, and can get ideal prediction result.
As a kind of prediction technique based on data mining, the present invention is using the historical time sequence of amount to be predicted as training
Data can make full use of the wide area measurement information in wind power plant；Meanwhile model considers adjacent node pair in learning process
The influence effectiveness of node to be predicted, therefore can reflect the space time correlation characteristic of electric system, to more accurately describe quantity of state
Changing rule and operation situation, reduce prediction error, obtain more structurally sound prediction result.
As a kind of dynamic modelling method, the present invention is not limited by preset parameter equation, can be according to the change of quantity of state
Change trend is constantly adjusted model, so that model has very strong adaptability.Therefore, under steady state conditions, model
It can be automatically adjusted to optimum state according to the variation of data, to ensure that high-precision prediction result.
As a kind of probability forecasting method, the present invention is on the basis for realizing the single desired value prediction of wind power plant future generated energy
On, also reliable prediction can be carried out for the waving interval of prediction error, is dispatching of power netwoks and electric system under specified confidence level
Reliability service more comprehensively predictive information is provided.
According to the Yantai Peng Lai wind power historical data grasped, sample carries out future 5 according to Various Seasonal respectively
Minute, 15 minutes futures, and prediction in following 30 minutes, prediction result are as shown in Figure 5 and Figure 6.Fig. 5 (a) is 5 minutes
Forecast interval, the forecast interval that Fig. 5 (b) is 15 minutes, the forecast interval that Fig. 5 (c) is 30 minutes, prediction season are autumn, number
According to temporal resolution be 5 minutes.As seen from Figure 5, as the increase of prediction scale, forecast interval constantly broaden, this is
As caused by the enhancing of wind-powered electricity generation uncertainty.Fig. 6 (a) and Fig. 6 (b) is the prediction probability distribution in summer and winter, pre- measurement ruler respectively
Degree is 15 minutes.As seen from Figure 6, the resulting prediction probability distribution of the method for the present invention can be well comprising true power output
Curve further demonstrates the validity of this method.
By comparing relative prediction residual with lasting method, management loading, Density Estimator equiprobability prediction technique
Index, the validity and practicability of the ultrashort-term wind power probability forecasting method proposed by the present invention based on experience dynamic modeling
It has been verified.
Fig. 7 is the structural representation of the ultrashort-term wind power probabilistic forecasting system of the invention based on experience dynamic modeling
Figure.
As shown in fig. 7, the ultrashort-term wind power probabilistic forecasting system of the invention based on experience dynamic modeling, comprising:
(1) non-linear degree of polymerization computing module is used to treat premeasuring time series and carries out standard normal processing, and
Data after standard normal are carried out with the calculating of the non-linear degree of polymerization, to investigate the nonlinear degree of given dynamical system.
Wherein, standard normal is that amount time series to be predicted is converted to the time series for obeying standardized normal distribution,
Even if it is desired for 0, variance 1:
Vt'=(Vt-μ(Vt))/σ(Vt) (20)
In formula, VtFor the former time series of amount to be predicted, μ (Vt) be the time series expectation, σ (Vt) it is the time sequence
The standard deviation of column, Vt' it is time series after standard normal.
Wherein, in the non-linear degree of polymerization computing module, to the data after standard normal using the method for S mapping
To calculate its non-linear degree of polymerization.S mapping calculation be lag coordinate vector and target variable between locally linear embedding,
It include a Dynamic gene θ, the factor be used to control the weight contacted between each vector: when θ=0 S be mapped as it is linear from
Regression model；And θ > 0 then imparts the more weights of adjacent states amount when calculating locally linear embedding, to show non-
Linearly.
(2) particle group optimizing module calculates optimal embedding dimension E and delay time T using particle swarm algorithm.
(3) higher-dimension phase space reconfiguration module is used for according to optimal embedding dimension E and delay time T, when treating premeasuring
Between sequence carry out phase space reconfiguration.
In the phase space reconfiguration module, given according to Ta Kensi theorem when optimal embedding dimension E meets E >=2M+1
Fixed dynamical system is reconstructed by the timing observation of one-dimensional observed quantity；Wherein, M is motive power system dimension.
Phase space reconfiguration, which refers to, is reconstructed former nonlinear dynamic system with the timing observation of one-dimensional, to describe it
Evolution and developing state.According to Ta Kensi theorem, as long as optimal embedding dimension E meets E >=2M+1, (M is motive power system
Dimension) when, given dynamical system can be reconstructed by the timing observation of one-dimensional observed quantity.Therefore, as long as choosing suitably
Optimal embedding dimension E and delay time T, so that it may in motive power system reconfiguration to a higher-dimension phase space, reconstruct
The analysis to original system is realized in space.
(4) prediction module is used to construct empirical dynamic model, carries out in phase space reconstruction to given dynamical system
Prediction, obtains the prediction result of amount to be predicted.
In the prediction module, the motion profile of quantity of state consecutive points is calculated using simplex sciagraphy, and to every
The probability that track occurs is estimated, and then predicts the motion profile of given dynamic system states.
Simplex sciagraphy is to be embedded into time delay in one single time series to generate attractor reconstruction, is led to
The probability of occurrence of the motion profile of calculating quantity of state consecutive points is crossed to predict that wind-powered electricity generation future time instance power probability is distributed.
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (22)
1. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling, which is characterized in that including following step
It is rapid:
Step (1), the nonlinear degree that given dynamical system is investigated based on amount time series data to be predicted；
Step (2), the optimal embedding dimension E and delay time T for calculating amount time series to be predicted；
Step (3), according to optimal embedding dimension E and delay time T, treat premeasuring time series and carry out phase space reconfiguration；
Step (4), building empirical dynamic model, predict given dynamical system in phase space reconstruction, obtain to pre-
The prediction result of measurement.
2. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as described in claim 1, special
Sign is that the step (1) specifically includes:
Step (1.1) treats premeasuring time series data and carries out standard normal processing；
Step (1.2) carries out non-linear degree of polymerization calculating to data after standard normal.
3. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as described in claim 1, special
Sign is, the non-linear degree of polymerization is calculated using S reflection method in the step (1.2).
4. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as described in claim 1, special
Sign is, calculates optimal embedding dimension E and the delay of amount time series to be predicted in step (2) using particle swarm optimization algorithm
Time τ；Wherein, E meets E >=2M+1, and M is motive power system dimension.
5. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 4, special
Sign is that the optimization aim of particle swarm optimization algorithm is to make to cover bandwidth index (CWC) minimum.
6. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as described in claim 1, special
Sign is, in step (3), the premeasuring time series for the treatment of carries out the process of phase space reconfiguration to obtain according to step (2)
Optimal embedding dimension E and delay time T given former dynamical system is reconfigured in a higher-dimension phase space.
7. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 6, special
Sign is, the model expression of the reconstruct higher-dimension phase space are as follows:
Xt=< Vt>
Wherein, XtFor based on amount timing observation { x to be predictedtConstruction E dimensional vector, Xt=< xt,xt-τ,xt-2τ,...,
xt-(E-1)τ>, t is to measure time, VtFor wind power observation sequence.
8. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as described in claim 1, special
Sign is, in step (4), during prediction in phase space reconstruction given dynamical system, is thrown using simplex
Shadow method calculates the motion profiles of quantity of state consecutive points, and the probability occurred to every track is estimated, and then predict given
Dynamic system states motion profile.
9. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 8, special
Sign is, in the simplex sciagraphy, quantity of state XsConsecutive points Xn(s,i)Position X after h time stepn(s,i)+hIt is
Quantity of state XsPosition X after h time steps+hProbability P (i) are as follows:
Wherein,
d(Xs,Xn(s,i)) it is quantity of state XsWith quantity of state Xn(s,i)Between Euclidean distance；N is the number of consecutive points, N=b, b=E
+1.Power probability distribution table according to each future value and its probability of appearance, after available h step-length；Further, when
When consecutive points quantity is sufficiently large, so that it may which approximation obtains the power probability distribution after h time step.
10. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 9, special
Sign is, further includes that the upper and lower area of prediction power is obtained from power probability distribution according to confidence level selected in advance
Between.
11. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 10,
It is characterized in that, further includes that the different confidence level of selection obtains the upper and lower section of different power, obtain the confidence of different in width
Band.
12. a kind of ultrashort-term wind power probabilistic forecasting system based on experience dynamic modeling characterized by comprising
Non-linear degree of polymerization computing module, for calculating the non-linear extent of polymerization of given dynamical system；
Optimal embedding dimension E and delay time T computing module；
Higher-dimension phase space reconfiguration module；And
Prediction module.
13. a kind of ultrashort-term wind power probabilistic forecasting system based on experience dynamic modeling as claimed in claim 12,
It is characterized in that, the non-linear degree of polymerization computing module further includes standard normalization processing submodule, for calculating given move
Before the non-linear degree of polymerization of state system, treats premeasuring time series and carry out standard normal processing.
14. a kind of ultrashort-term wind power probabilistic forecasting system based on experience dynamic modeling as claimed in claim 12,
It is characterized in that, in the non-linear degree of polymerization computing module, the non-linear of the given dynamical system is calculated using S reflection method
The degree of polymerization.
15. a kind of ultrashort-term wind power probabilistic forecasting system based on experience dynamic modeling as claimed in claim 12,
It is characterized in that, after completing the non-linear degree of polymerization calculating of given dynamical system, amount to be predicted is calculated using particle swarm optimization algorithm
The optimal embedding dimension E and delay time T of time series；Wherein, E meets E >=2M+1, and M is motive power system dimension.
16. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 12,
It is characterized in that, the optimization aim of particle swarm optimization algorithm is to make to cover bandwidth index (CWC) minimum.
17. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 12,
It is characterized in that, the former dynamical system that the higher-dimension phase space reconfiguration module will be given based on optimal embedding dimension E and delay time T
It is reconfigured in a higher-dimension phase space.
18. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 17,
It is characterized in that, the model expression of the higher-dimension phase space are as follows:
Xt=< Vt>
Wherein, XtFor by timing observation { xtConstruction E dimensional vector, Xt=< xt,xt-τ,xt-2τ,...,xt-(E-1)τ>, VtFor wind
Electric field power observation sequence.
19. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 18,
It being characterized in that, the higher-dimension phase space reconfiguration module calculates the motion profile of quantity of state consecutive points using simplex sciagraphy,
And the probability occurred to every track is estimated, and then predicts the motion profile of given dynamic system states.
20. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 19,
It is characterized in that, in the simplex sciagraphy, quantity of state XsConsecutive points Xn(s,i)Position X after h time stepn(s,i)+h
It is quantity of state XsPosition X after h time steps+hProbability P (i) are as follows:
Wherein, d (Xs,Xn(s,i)) it is quantity of state XsWith quantity of state Xn(s,i)Between Euclidean distance；N is the number of consecutive points, N=
B, b=E+1.Power probability distribution table according to each future value and its probability of appearance, after available h step-length；Into one
Step ground, when consecutive points quantity is sufficiently large, so that it may which approximation obtains the power probability distribution after h time step.
21. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 20,
It is characterized in that, further includes that the upper and lower of prediction power is obtained from power probability distribution according to confidence level selected in advance
Section.
22. a kind of ultrashort-term wind power probability forecasting method based on experience dynamic modeling as claimed in claim 21,
It is characterized in that, further includes that the different confidence level of selection obtains the upper and lower section of different power, obtain the confidence of different in width
Band.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910009512.7A CN109886452B (en) | 2019-01-04 | 2019-01-04 | Wind power ultra-short term probability prediction method and system based on empirical dynamic modeling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910009512.7A CN109886452B (en) | 2019-01-04 | 2019-01-04 | Wind power ultra-short term probability prediction method and system based on empirical dynamic modeling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886452A true CN109886452A (en) | 2019-06-14 |
| CN109886452B CN109886452B (en) | 2021-06-15 |
Family
ID=66925662
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910009512.7A Active CN109886452B (en) | 2019-01-04 | 2019-01-04 | Wind power ultra-short term probability prediction method and system based on empirical dynamic modeling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886452B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110188967A (en) * | 2019-06-21 | 2019-08-30 | 合肥工业大学 | Electric load probabilistic forecasting method based on chaos crowd algorithm and Bayesian network |
| CN110264006A (en) * | 2019-06-21 | 2019-09-20 | 合肥工业大学 | Wind power probabilistic forecasting method based on chaos glowworm swarm algorithm and Bayesian network |
| CN111445010A (en) * | 2020-03-26 | 2020-07-24 | 南京工程学院 | Distribution network voltage trend early warning method based on evidence theory fusion quantum network |
| CN111914479A (en) * | 2020-07-03 | 2020-11-10 | 天津大学 | Grouting power mixed interval prediction method for dam bedrock |
| CN112653199A (en) * | 2020-12-17 | 2021-04-13 | 华中科技大学 | Multi-wind-farm joint output prediction method and device based on dynamic R-Vine Copula model |
| CN111915083B (en) * | 2020-08-03 | 2024-06-11 | 国网山东省电力公司电力科学研究院 | Wind power prediction method and prediction system based on time layered combination |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880989A (en) * | 2012-09-05 | 2013-01-16 | 中国电力科学研究院 | Method for modeling wind power output time sequence |
| CN103268519A (en) * | 2013-04-26 | 2013-08-28 | 哈尔滨工程大学 | Electric power system short-term load forecast method and device based on improved Lyapunov exponent |
| CN107067118A (en) * | 2017-05-17 | 2017-08-18 | 山东大学 | A kind of electrical quantity ultra-short term prediction method and system based on data mining |
-
2019
- 2019-01-04 CN CN201910009512.7A patent/CN109886452B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880989A (en) * | 2012-09-05 | 2013-01-16 | 中国电力科学研究院 | Method for modeling wind power output time sequence |
| CN103268519A (en) * | 2013-04-26 | 2013-08-28 | 哈尔滨工程大学 | Electric power system short-term load forecast method and device based on improved Lyapunov exponent |
| CN107067118A (en) * | 2017-05-17 | 2017-08-18 | 山东大学 | A kind of electrical quantity ultra-short term prediction method and system based on data mining |
Non-Patent Citations (3)
| Title |
|---|
| JETHRO DOWELL等: "Very-Short-Term Probabilistic Wind Power Forecasts by Sparse Vector Autoregression", 《IEEE TRANSACTIONS ON SMART GRID》 * |
| 关文渊: "风电功率预测不确定性分析及应用研究", 《中国优秀硕士学位论文全文数据库》 * |
| 朱霄珣等: "基于PSO参数优化的LS-SVM风速预测方法研究", 《中国电机工程学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110188967A (en) * | 2019-06-21 | 2019-08-30 | 合肥工业大学 | Electric load probabilistic forecasting method based on chaos crowd algorithm and Bayesian network |
| CN110264006A (en) * | 2019-06-21 | 2019-09-20 | 合肥工业大学 | Wind power probabilistic forecasting method based on chaos glowworm swarm algorithm and Bayesian network |
| CN110264006B (en) * | 2019-06-21 | 2021-01-15 | 合肥工业大学 | Wind power probabilistic prediction method based on chaotic firefly algorithm and Bayesian network |
| CN110188967B (en) * | 2019-06-21 | 2021-01-15 | 合肥工业大学 | Power load probabilistic prediction method based on chaos crowd algorithm and Bayesian network |
| CN111445010A (en) * | 2020-03-26 | 2020-07-24 | 南京工程学院 | Distribution network voltage trend early warning method based on evidence theory fusion quantum network |
| CN111445010B (en) * | 2020-03-26 | 2021-03-19 | 南京工程学院 | Distribution network voltage trend early warning method based on evidence theory fusion quantum network |
| CN111914479A (en) * | 2020-07-03 | 2020-11-10 | 天津大学 | Grouting power mixed interval prediction method for dam bedrock |
| CN111915083B (en) * | 2020-08-03 | 2024-06-11 | 国网山东省电力公司电力科学研究院 | Wind power prediction method and prediction system based on time layered combination |
| CN112653199A (en) * | 2020-12-17 | 2021-04-13 | 华中科技大学 | Multi-wind-farm joint output prediction method and device based on dynamic R-Vine Copula model |
| CN112653199B (en) * | 2020-12-17 | 2022-08-02 | 华中科技大学 | Multi-wind-farm joint output prediction method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886452B (en) | 2021-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886452A (en) | A kind of ultrashort-term wind power probability forecasting method and system based on experience dynamic modeling | |
| CN103390116B (en) | Use the photovoltaic power station power generation power forecasting method of stepping mode | |
| An et al. | Short-term wind power prediction based on particle swarm optimization-extreme learning machine model combined with AdaBoost algorithm | |
| Flores et al. | Application of a control algorithm for wind speed prediction and active power generation | |
| Abid et al. | Optimal planning of multiple renewable energy-integrated distribution system with uncertainties using artificial hummingbird algorithm | |
| CN105046374A (en) | Power interval predication method based on nucleus limit learning machine model | |
| CN103400009B (en) | Based on the wind energy turbine set Dynamic Equivalence of the semi-supervised spectral clustering of division level | |
| CN108388962B (en) | Wind power prediction system and method | |
| CN102102626A (en) | Method for forecasting short-term power in wind power station | |
| CN105512745A (en) | Wind power section prediction method based on particle swarm-BP neural network | |
| CN106875033A (en) | A kind of wind-powered electricity generation cluster power forecasting method based on dynamic self-adapting | |
| Wu et al. | Stratification-based wind power forecasting in a high-penetration wind power system using a hybrid model | |
| CN101794996A (en) | Real-time predicting method for output of wind electric field | |
| Rizwan et al. | Artificial intelligence based approach for short term load forecasting for selected feeders at madina saudi arabia | |
| CN103683274A (en) | Regional long-term wind power generation capacity probability prediction method | |
| CN105846425A (en) | Economic dispatching method based on general wind power forecasting error model | |
| CN111612244B (en) | QRA-LSTM-based method for predicting nonparametric probability of photovoltaic power before day | |
| CN106651007A (en) | Method and device for GRU-based medium and long-term prediction of irradiance of photovoltaic power station | |
| CN112968441A (en) | Power grid planning method applied to large-scale wind power base | |
| Aliberti et al. | Forecasting Short-term Solar Radiation for Photovoltaic Energy Predictions. | |
| CN110414725A (en) | The integrated wind power plant energy-storage system dispatching method of forecast and decision and device | |
| Nagaraja et al. | A survey on wind energy, load and price forecasting:(Forecasting methods) | |
| Bo et al. | Hybrid PSO-BP neural network approach for wind power forecasting | |
| Pappala et al. | Predictive optimal control of wind farm reactive sources | |
| CN107194507A (en) | A kind of short-term wind speed forecasting method of wind farm based on combination SVMs |
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 |