CN109376897A - A kind of short-term wind power forecast method based on hybrid algorithm - Google Patents
A kind of short-term wind power forecast method based on hybrid algorithm Download PDFInfo
- Publication number
- CN109376897A CN109376897A CN201810997267.0A CN201810997267A CN109376897A CN 109376897 A CN109376897 A CN 109376897A CN 201810997267 A CN201810997267 A CN 201810997267A CN 109376897 A CN109376897 A CN 109376897A
- Authority
- CN
- China
- Prior art keywords
- component
- imf
- decomposition
- wind power
- empirical mode
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 105
- 230000006870 function Effects 0.000 claims abstract description 11
- 239000000284 extract Substances 0.000 claims abstract description 9
- 238000010183 spectrum analysis Methods 0.000 claims abstract description 9
- 230000000717 retained effect Effects 0.000 claims abstract description 3
- 238000012549 training Methods 0.000 claims description 18
- 230000005611 electricity Effects 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000005516 engineering process Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000007423 decrease Effects 0.000 claims description 4
- 238000004519 manufacturing process Methods 0.000 claims description 4
- 102000002274 Matrix Metalloproteinases Human genes 0.000 claims description 3
- 108010000684 Matrix Metalloproteinases Proteins 0.000 claims description 3
- 230000003416 augmentation Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Public Health (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Wind Motors (AREA)
Abstract
The present invention relates to a kind of short-term wind power forecast methods based on hybrid algorithm, the following steps are included: original wind power is decomposed into a series of intrinsic mode functions (IMF) sub- modal components using integrated Empirical mode decomposition by S1, S2 is decomposed integrated Empirical mode decomposition using singular spectrum analysis method resulting in addition to first IMF component IMF1Except each IMF component and the main trend component of RES component extract, to obtain the sub- modal components that sequence signature becomes apparent, S3 is to IMF1Residual components R obtained in component and S2 is retained, and to IMF1Component and residual components R carry out WAVELET PACKET DECOMPOSITION, obtain a series of more stable new submodule states, using online robust extreme learning machine, all submodule states obtained to S1-S3 step establish prediction model to S4 respectively, and final wind power prediction result is obtained by superposition;The present invention can carry out effective Accurate Prediction to practical wind power system, provide important references for the operation and planning of electric system.
Description
Technical field
The present invention relates to electric system prediction technique fields, more particularly, to a kind of short-term wind based on hybrid algorithm
Electrical power prediction technique.
Background technique
With the development of wind-power electricity generation, the uncertain stability to electric system and electricity market of wind power is filled
The influence of abundant property and economy is also increasingly shown.Thus, the prediction of accurate short-term wind-electricity power to the planning of electric system and
Scheduling is of great significance.Currently, wind-powered electricity generation prediction technique is broadly divided into statistical learning method according to the data source difference used
And physical method.Wherein, statistical learning method establishes statistical learning according to wind power plant historical measurement data and periphery measurement data
Model, most common Statistical learning model include Time Series Analysis Model, artificial nerve network model, support vector machines mould
The machine learning methods such as type (supportvectormachine, SVM).Traditional artificial neural network and support vector machines knot
Structure is complex, and parameter is various, and extreme learning machine (extremelearningmachine, ELM) is single hidden layer feed forward neural
Network has the characteristics that pace of learning is fast, generalization is good.In addition, improved ELM model, which is also put forward one after another, is applied to prediction
Field, such as: robust extreme learning machine (outlierrobust extremelearningmachine, ORELM) compared to ELM and
Speech can better adapt to the outlier being likely to occur in training set, online extreme learning machine (onlinesequential
Extremelearningmachine, OSELM) line mechanism that has can be realized in actual time-varying system it is preferably pre-
Survey effect.
Currently, data decomposition technique has been widely used in short-term wind-electricity power prediction, the clock synchronization in the way of effective
Between sequence carry out decomposition pretreatment, for capture data characteristic rule and improve precision of prediction all play an important role.
The data resolving method of more mainstream have experience mode decomposition technology (empirical mode decomposition, EMD),
Integrated Empirical mode decomposition (ensemble empirical mode decomposition, EEMD), become mode decomposition
Technology (variational mode decomposition, VMD) and WAVELET PACKET DECOMPOSITION technology (wavelet packet
Decomposition, WPD) etc..
Currently used short-term wind-electricity power prediction scheme mainly has:
(1) the short-term wind-electricity power prediction model of empirical mode decomposition (EMD) and extreme learning machine (ELM) are based on:
First with empirical mode decomposition (EMD) by original wind-powered electricity generation Time Series be a series of submodule state, then
ELM model is established to each subcomponent and is predicted, obtains final prediction result finally by superposition.
(2) based on integrated empirical mode decomposition (EEMD)-principal component analysis (principalcomponentanalysis,
) and the combination forecasting of support vector machines (SVM) PCA:
Decomposed to obtain one group not to nonstationary time series first with integrated Empirical mode decomposition (EEMD)
With the sub- modal components of scale, noise present in each submodule state then is removed using PCA, reduces intrinsic dimensionality and redundancy
Degree, finally establishes SVM model and predicts principal component as input variable.
Currently used short-term wind-electricity power prediction scheme has the disadvantage that
(1) EMD is decomposed very sensitive to sampling mode and noise, and integrated Empirical mode decomposition (EEMD) passes through benefit
The statistical property (frequency distribution obedience is uniformly distributed) having with zero mean Gaussian white noise sequence compensates for EMD when decomposing
It is easy to cause the defect of modal aliasing.
(2) traditional neural network, such as typical BP neural network, parameter is many and diverse, and rate of convergence is very slow and is easily trapped into
Locally optimal solution;Original extreme learning machine convergence rate quickly, but is susceptible to the interference of the outlier in training set, and
Improved robust extreme learning machine can preferably adapt to the case where sample contains outlier.However, actual wind power category
In the on-line system of time-varying, above prediction model is still undesirable for the prediction effect of practical time-varying system.
Summary of the invention
The present invention is to overcome defect described in the above-mentioned prior art, provides a kind of short-term wind-electricity function based on hybrid algorithm
Rate prediction technique.
It the described method comprises the following steps:
S1: original wind power is decomposed into a series of eigen mode letters using integrated Empirical mode decomposition (EEMD)
Number (intrinsic mode function, IMF) sub- modal components, i.e. IMF component;
S2: using singular spectrum analysis (SSA) method by integrated Empirical mode decomposition (EEMD) decompose it is resulting in addition to
First IMF component IMF1Except each IMF component and the main trend component of residual components RES extract, to obtain sequence
The sub- modal components that column feature becomes apparent, and the residual components R after being extracted;
S3: in order to avoid directly rejecting IMF1Component and singular spectrum analysis (SSA) extract the remainder after main trend component
And the loss of signal detail information is caused, precision of prediction reduction is in turn resulted in, to IMF1Component and residual components R are retained,
And to IMF1Component and residual components R carry out WAVELET PACKET DECOMPOSITION, obtain a series of more stable new submodule states;
S4: using online robust extreme learning machine, all submodule states obtained to S1-S3 step establish prediction mould respectively
Type, and final wind power prediction result is obtained by superposition.
The invention proposes a kind of short-term wind power forecast methods based on hybrid algorithm, wherein first with integrated warp
Mode decomposition technology (EEMD) is tested original wind power time series is decomposed to obtain a series of IMF modal components, then
Obtain the main trend sequence of each IMF component (in addition to IMF1) using singular spectrum analysis (SSA), then to IMF1 component and
Residual components R carries out WAVELET PACKET DECOMPOSITION to obtain more stable new submodule state;Compared with single decomposition method, have more preferable
Discomposing effect, can preferably capture sequence signature, obtain the smaller submodule state of more stable fluctuation, be conducive to improve prediction
Precision;The online robust extreme learning machine (OSORELM) used in the present invention has higher precision of prediction, and can be preferably
It is adapted to actual time-varying system.
Preferably, the step S1 specifically includes the following steps:
S1.1: N is separately added into original wind-powered electricity generation time series x (t)GSecondary mean value is zero, and amplitude standard deviation is the height of constant
This white noise ωi(t), new wind-powered electricity generation time series x is obtainedi(t):
xi(t)=x (t)+ωi(t), i=1,2 ..., NG
Wherein, NGFor the integer greater than 2;ωi(t) intensity depends on white Gaussian noise standard deviation and initial signal standard
The ratio between difference.
S1.2: to each signal xi(t) Empirical mode decomposition (empirical mode is carried out respectively
Decomposition, EMD) decompose, obtain m intrinsic mode functions (intrinsic mode function, IMF) component with
One residual components Ri(t):
Wherein, imfijIt (t) is that i-th is added white Gaussian noise and decomposes obtain the through Empirical mode decomposition (EMD)
J IMF component;
S1.3: to all IMF component imfij(t) mean value computation is carried out, final IMF component IMF is obtainedj(t) and it is remaining
Components R ES (t):
Wherein, IMFjIt (t) is integrated Empirical mode decomposition (ensemble empirical mode
Decomposition, EEMD) decompose j-th obtained of IMF component;
S1.4: according to above-mentioned steps S1.1-S1.3, by integrating Empirical mode decomposition (ensemble empirical
Mode decomposition, EEMD) by original wind power Time Series be a series of IMF modal components.
Preferably, the step S2 specifically includes the following steps:
S2.1: it decomposes: selected length of window L, whereinBy wind-powered electricity generation time series X=(x1,x2,…,xN), i.e.,
Respectively S1 decomposes the multi-dimensional matrix that resulting each IMF component is converted into L × M:
Wherein, M=N-L+1;
S2.2: it calculates: calculating GGT, and carry out singular value decomposition and obtain L characteristic value: λ1≥λ2≥…≥λL>=0, often
A characteristic value represents corresponding singular value decomposition (SVD) component and contributes the energy of original signal, the corresponding spy of each characteristic value
Levying vector is respectively U1,U2,…UL, and:
G=G1+G2+…Gd
Wherein, GiIndicate i-th of singular value decomposition (SVD) component,And UiFor
Ith feature is worth corresponding feature vector, d=max { i: λi> 0 };
S2.3: reconstruct: singular value decomposition (SVD) component of condition needed for selection meets constitutes a subset { GI,It is available:
Wherein, λjFor j-th of characteristic value, UjFor the corresponding feature vector of j-th of characteristic value;
For GIContribution rate, and by the matrix G of L × MIIt is reconstructed into the main trend component Y=that length is N
(y1,y2,...,yN);
Enable L*=min (L, M), K*=max (L, M), if L≤M,It is on the contrary thenPass through following formula
Calculate reproducing sequence:
Wherein, gm,nRepresenting matrixIn m row n-th arrange element;
S2.4: according to step S2.1-S2.3, integrated Empirical mode decomposition (EEMD) is decomposed resulting in addition to the
The main trend component of each IMF component and RES component except one IMF1 component extracts, and respectively corresponds to obtain one
A remainder rk;
The remainder of each IMF component is superimposed the residual components R total as one to handle, it may be assumed that
Wherein, rkCorrespond to the remainder of k-th of IMF component.
Preferably, the singular value for reconstructing each main trend component of IMF component how is selected in the step S2.3
Decompose (SVD) component subset { GI, concrete operations are as follows:
It is located at highest frequency range, institute since integrated Empirical mode decomposition (EEMD) decomposes resulting first IMF component
The noise energy contained is most, so usually all being handled as high frequency noise, the noise energy in remaining each IMF component
As the increase of order is constantly successively decreased, production decline law is as follows:
Wherein, EnkFor the energy of institute's Noise in k-th of IMF component, δ ≈ 0.719, ε ≈ 2.01;
The gross energy of each component can indicate are as follows:
Wherein, imfkIt (i) is i-th of element of k-th of IMF component.
By En1For the 1st IMF component IMF1The energy of middle institute's Noise, by IMF1All as high frequency noise, can be obtained
En1=E1, energy shared by signal in remaining IMF component (k >=2) are as follows:
Exk=Ek-Enk
So IMFkRatio shared by signal energy can calculate in (k >=2) are as follows:
Rk=Exk/Ek
On this basis, selection reconstruct Shi Neng represents input signal IMFkSingular value decomposition (SVD) component of (k >=2),
The specific method is as follows:
Work as Rk>=0.5, illustrate that signal is main component in the IMF component, by Exk=Ek-EnkCalculate H:
At this point, can represent this IMF sequence singular value decomposition (SVD) component constitute subset as
Work as Rk≤ 0.5, illustrate that noise is main component in the IMF component, by Rk=Exk/EkCalculate H:
At this point, can represent this IMF sequence singular value decomposition (SVD) component constitute subset as
Preferably, the step S3 specifically includes the following steps:
S3.1: choosing the parameter of WAVELET PACKET DECOMPOSITION, sets three layers for Decomposition order, will using wavelet packet decomposition algorithm
The biggish submodule state of got in step 2 SE value is decomposed, and obtains one three layers of wavelet packet tree, form is binary tree;
S3.2: wavelet packet tree bottom node, i.e., the signal of eight nodes, so that it may after obtaining double decomposition are read respectively
New submodule state sequence;
S3.3: IMF is respectively obtained1Two groups after WAVELET PACKET DECOMPOSITION technology (WPD) decomposition of component and residual components R
New submodule state sequence.
Preferably, the step S4 specifically includes the following steps:
S4.1: initial phase utilizes initial training collectionEstablish initial robust extreme learning machine
Model, constraint condition are as follows:
β0Weight, e are initially exported for hidden layer0For the training error of initial training collection, N0For initial training collection sample size,
C is regularization coefficient;H0For initial hidden layer output matrix, H0It can calculate are as follows:
Wherein, Q is node in hidden layer;
The constraint condition optimization problem can be iterated calculating by augmentation Lagrangian, its calculation formula is:
Wherein, μ=2N0/||Y0||1For penalty factor, λkFor the Lagrangian in kth time iteration;
According toAvailable βk+1And ek+1Final expression formula are as follows:
Wherein, I is unit matrix;
S4.2: the on-line study stage, when+1 sample of kth of training set D arrives, using recurrent least square method come
It updates and calculates hidden layer output weight, calculation formula are as follows:
Wherein, hk+1=[g (w1·xi+b1)…g(wQ·xi+bQ)];Error is predicted for new samples;ζk+1And εk+1It is
Assist parameter, wherein work as ζk+1When=0, then Pk+1=Pk;
S4.3: initialization forgetting factor θ is 0≤θ0≤ 1, and update is iterated to θ according to the following formula:
vk+1=θk(vk+1)
Wherein, νk+1, ηk+1, γk+1It is auxiliary parameter, ρ is a positive number, and the value range of the initial value of ν, γ is
[0,1];
Relevant parameter may be configured as: T0=20, θ0=1, ν0=10-6, γ0=10-3, ρ=0.99.
Compared with prior art, the beneficial effect of technical solution of the present invention is: the present invention have better discomposing effect,
Higher precision of prediction, faster calculating speed can be preferably adapted in the time-varying system in practical application.
Detailed description of the invention
Fig. 1 is the short-term wind power forecast method flow chart based on hybrid algorithm.
Fig. 2 is in January, 2017 original wind power time series chart.
Fig. 3 is in March, 2017 original wind power time series chart.
Fig. 4 is in June, 2017 original wind power time series chart.
Fig. 5 is the result figure for carrying out integrated Empirical mode decomposition to original wind power and decomposing.
Fig. 6 is the main trend component result figure of each IMF component and RES component.
Fig. 7 is the result figure that wavelet decomposition is carried out to IMF1 component.
Fig. 8 is the result figure that wavelet decomposition is carried out to residual components R.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent;
In order to better illustrate this embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent practical production
The size of product;
To those skilled in the art, it is to be understood that certain known features and its explanation, which may be omitted, in attached drawing
's.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
The present invention is a kind of short-term wind power forecast method based on hybrid algorithm, implementation flow chart such as Fig. 1 institute
Show, the specific steps of technical solution of the present invention are as follows:
S1. original wind power is decomposed into a series of submodule states using integrated Empirical mode decomposition (EEMD), had
Body step are as follows:
S1.1: N is separately added into original wind-powered electricity generation time series x (t)GSecondary mean value is zero, and amplitude standard deviation is the height of constant
This white noise ωi(t), new wind-powered electricity generation time series x is obtainedi(t):
xi(t)=x (t)+ωi(t), i=1,2 ..., NG
Wherein, ωi(t) intensity depends on the ratio between white Gaussian noise standard deviation and initial signal standard deviation, the present embodiment
Middle setting Nstd=0.5, NG=200.
S1.2: to each signal xi(t) EMD decomposition is carried out respectively, obtains m intrinsic mode functions (intrinsic mode
Function, IMF) component and a residual components Ri(t):
Wherein, imfijIt (t) is that i-th addition white Gaussian noise warp (EMD) decomposes j-th obtained of IMF component.
S1.3: to all IMF component imfij(t) mean value computation is carried out, final IMF component IMF is obtainedj(t) and it is remaining
Components R ES (t):
Wherein, IMFj(t) j-th obtained of IMF component is decomposed for integrated Empirical mode decomposition (EEMD).
S1.4:, can be by integrating Empirical mode decomposition (EEMD) for original wind-powered electricity generation function according to above-mentioned steps S11-S13
Rate Time Series are a series of IMF modal components.
S2: extracting the main trend component in each IMF component using singular spectrum analysis (SSA) method, to obtain sequence spy
Levy the sub- modal components become apparent, the specific steps are as follows:
S2.1: it decomposes: selected length of window LBy wind-powered electricity generation time series X=(x1,x2,…,xN) (distinguish
Resulting each IMF component is decomposed for previous step) it is converted into the multi-dimensional matrix of L × M:
Wherein, M=N-L+1 calculates GGT, and carry out singular value decomposition and obtain L characteristic value: λ1≥λ2≥…≥λL≥
0, each characteristic value represents corresponding singular value decomposition (SVD) component and contributes the energy of original signal.Each characteristic value is corresponding
Feature vector be respectively U1,U2,…UL, and:
G=G1+G2+…Gd
Wherein, GiIndicate i-th of singular value decomposition (SVD) component,And UiFor
Ith feature is worth corresponding feature vector, d=max { i: λi> 0 };
S2.2: reconstruct: singular value decomposition (SVD) component of condition needed for selection meets constitutes a subset { GI,It is available:
Wherein, λjFor j-th of characteristic value, UjFor the corresponding feature vector of j-th of characteristic value;
Wherein,For GIContribution rate, and by the matrix G of L × MIIt is reconstructed into the main trend point that length is N
Measure Y=(y1,y2,...,yN)。
Enable L*=min (L, M), K*=max (L, M), if L≤M,It is on the contrary thenPass through following formula
Calculate reproducing sequence:
Wherein, gm,nRepresenting matrixIn m row n-th arrange element;
S2.3: where how to select singular value decomposition (SVD) component for reconstructing each main trend component of IMF component
Subset { GI, concrete operations are as follows:
It is located at highest frequency range, institute since integrated Empirical mode decomposition (EEMD) decomposes resulting first IMF component
The noise energy contained is most, so usually all being handled as high frequency noise, the noise energy in remaining each IMF component
As the increase of order is constantly successively decreased, production decline law is as follows:
Wherein, EnkFor the energy of institute's Noise in k-th of IMF component, δ ≈ 0.719, ε ≈ 2.01;
The gross energy of each component can indicate are as follows:
Wherein, imfkIt (i) is i-th of element of k-th of IMF component;
According to above-mentioned theory, E can be obtainedn1=E1;Energy shared by signal in remaining IMF component (k >=2) are as follows:
Exk=Ek-Enk
So IMFkRatio shared by signal energy can calculate in (k >=2) are as follows:
Rk=Exk/Ek
On this basis, selection reconstruct Shi Neng represents input signal IMFkSingular value decomposition (SVD) component of (k >=2),
The specific method is as follows:
Work as Rk>=0.5, illustrate that signal is main component in the IMF component, by Exk=Ek-EnkCalculate H:
At this point, can represent this IMF sequence singular value decomposition (SVD) component constitute subset as
Work as Rk≤ 0.5, illustrate that noise is main component in the IMF component, by Rk=Exk/EkCalculate H:
At this point, can represent this IMF sequence singular value decomposition (SVD) component constitute subset as
S2.4: according to step S2.1-S2.3, integrated Empirical mode decomposition (EEMD) is decomposed resulting (in addition to the
Except one IMF component) the main trend component of each IMF component and RES component extracts, and respectively corresponds to obtain one
A remainder rk;The remainder of each IMF component is superimposed the residual components R total as one and handled by the present embodiment, it may be assumed that
Wherein, rkCorrespond to the remainder of k-th of IMF component.
S3: in order to avoid directly rejecting IMF1Component and singular spectrum analysis (SSA) extract the remainder after main trend component
And the loss of signal detail information is caused, precision of prediction reduction is in turn resulted in, therefore to IMF1Component and residual components R are protected
It stays, and to IMF1Component and residual components R carry out WAVELET PACKET DECOMPOSITION, obtain a series of more stable new submodule states, specific to walk
It is rapid as follows:
S3.1: choosing the parameter of WAVELET PACKET DECOMPOSITION, sets three layers for Decomposition order, will using wavelet packet decomposition algorithm
The biggish submodule state of got in step 2 SE value is decomposed, and one three layers of wavelet packet tree is obtained (form is binary tree);
S3.2: the signal of wavelet packet tree bottom node (namely eight nodes) is read respectively to get double decomposition is arrived
New submodule state sequence afterwards;
S3.3: IMF is respectively obtained1Two groups after WAVELET PACKET DECOMPOSITION technology (WPD) decomposition of component and residual components R
New submodule state sequence.
S4: using online robust extreme learning machine, all submodule states obtained to above-mentioned steps establish prediction mould respectively
Type, and final wind power prediction obtained by superposition as a result, the specific steps are that:
S4.1: initial phase.Utilize initial training collectionEstablish initial robust extreme learning machine
Model, constraint condition are
β0Weight, e are initially exported for hidden layer0For the training error of initial training collection, N0For initial training collection sample size,
C is regularization coefficient;H0For initial hidden layer output matrix, H0It can calculate are as follows:
Wherein, Q is node in hidden layer;
The constraint condition optimization problem can be iterated calculating by augmentation Lagrangian, its calculation formula is:
Wherein, μ=2N0/||Y0||1For penalty factor, λkFor the Lagrangian in kth time iteration;
According toAvailable βk+1And ek+1Final expression formula are as follows:
Wherein, I is unit matrix;
S4.2: on-line study stage.When+1 sample of kth of training set D arrives, using recurrent least square method come
It updates and calculates hidden layer output weight, calculation formula is
Wherein, hk+1=[g (w1·xi+b1)g(wQ·xi+bQ)];Error is predicted for new samples;ζk+1And εk+1Supplemented by
Help parameter, wherein work as ζk+1When=0, then Pk+1=Pk;
S4.3: initialization forgetting factor θ is θ≤θ0≤ 1, and update is iterated to θ according to the following formula:
vk+1=θk(vk+1)
Wherein, νk+1, ηk+1, γk+1It is auxiliary parameter, ρ is a positive number.The value range of the initial value of ν, γ is
[0,1];
The present embodiment will be to each parameter setting are as follows: T0=20, θ0=1, ν0=10-6, γ0=10-3, ρ=0.99.
The present embodiment chooses the actual measurement wind-powered electricity generation function in 2017 that Spain Sotavento Galicia wind power plant provides
Rate data (/ hour) are used as research object, take January, March, June to be analyzed as allusion quotation example respectively, as in Figure 2-4.
Wherein, data sampling point is a point per hour, takes 700 hours first (totally 700 points) therein as test sample, wherein preceding
600 points carry out 1 step in advance that input dimension is 6 and (it is small to shift to an earlier date 1 as training sample, rear 100 points as test sample
When) wind power prediction test.
For the performance of qualitatively valuation prediction models, introduce mean absolute error (mean absolute error,
) and Performance Evaluating Indexes of the root-mean-square error (root mean-squared error, RMSE) as prediction model MAE:
Wherein, yiWithIt is wind speed actual value and wind speed value respectively, N is the sample size of test set.
(1) it is with the wind power time series in March, 2017 (be divided between data sampling 1 hour, totally 744 points)
Example, is decomposed into a series of submodule states for original wind power using integrated Empirical mode decomposition (EEMD), decomposition result is such as
Shown in Fig. 5.
As seen from Figure 5, original wind power passes through a series of submodule states of integrated Empirical mode decomposition (EEMD)
In, IMF1Component be it is the most chaotic and unordered, illustrate IMF1Noise contained in component is most, is carrying out singular spectrum analysis
It (SSA) can be first by IMF when extracting main trend component1Component is all considered as noise processed.
(2) integrated Empirical mode decomposition (EEMD) is decomposed into resulting (remove using singular spectrum analysis (SSA) algorithm
IMF1Except component) the main trend component of each IMF component and RES component extracts, as a result as shown in Figure 6.
(3) to IMF1Component and residual components R carry out WAVELET PACKET DECOMPOSITION respectively, and obtain two groups of more stable fluctuations more
Small submodule state, is denoted as WPD10~WPD17 and WPD20~WPD27 respectively;Wherein, to IMF1Component carries out WAVELET PACKET DECOMPOSITION
As a result as shown in fig. 7, the result for carrying out WAVELET PACKET DECOMPOSITION to residual components R is as shown in Figure 8.
By Fig. 7 and Fig. 8 as it can be seen that IMF1Component and residual components R are obtaining one group respectively after WAVELET PACKET DECOMPOSITION
More stable submodule state sequence, and sequence signature becomes apparent.
(4) prediction basic mode type (OSORELM) and ELM, ORELM that this method proposes are compared, the error of each model
Evaluation index result difference is as shown in table 1.
The error assessment index result of each model of table 1
As known from Table 1, prediction basic mode type (OSORELM) precision of prediction that this method proposes is substantially better than other models.
(5) by this method (being denoted as EEMD-SSA-WPD-OSORELM) with decomposed based on EMD, EEMD, WPD and VMD it is pre-
It surveys model and compares experiment, the error assessment index result difference of each model is as shown in table 4.
The error assessment index result of each model of table 2
As known from Table 2, the estimated performance of the mixed method proposed based on this method is substantially better than other several decomposition sides
Method shows the superiority for the mixed decomposition method that this method is proposed.
(6) in addition, the wind power time series for choosing in June, 2017 and September again carries out further this method
Verifying, is utilized the error assessment index knot that EEMD-SSA-WPD-OSORELM predicts different month wind powers
Fruit difference is as shown in table 3.
The error assessment index result of the different month wind power predictions of table 3
As known from Table 3, this method is all higher to the precision of prediction in different months, and showing this method can preferably fit
The wind power prediction situation for answering different characteristics, can be well adapted for actual time-varying system, it was demonstrated that this method it is effective
Property.
The same or similar label correspond to the same or similar components;
The terms describing the positional relationship in the drawings are only for illustration, should not be understood as the limitation to this patent;
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (6)
1. a kind of short-term wind power forecast method based on hybrid algorithm, which is characterized in that the described method comprises the following steps:
S1: original wind power is decomposed into a series of intrinsic mode functions submodule states point using integrated Empirical mode decomposition
Amount, i.e. IMF component;
S2: using singular spectrum analysis method that the decomposition of integrated Empirical mode decomposition is resulting in addition to first IMF component IMF1
Except each IMF component and the main trend component of residual components RES extract, to obtain the son that sequence signature becomes apparent
Modal components, and the residual components R after being extracted;
S3: to IMF1Component and residual components R are retained, and to IMF1Component and residual components R carry out WAVELET PACKET DECOMPOSITION, obtain
To a series of more stable new submodule states;
S4: using online robust extreme learning machine, all submodule states obtained to S1-S3 step establish prediction model respectively, and
Final wind power prediction result is obtained by superposition.
2. the short-term wind power forecast method according to claim 1 based on hybrid algorithm, which is characterized in that described
Step S1 specifically includes the following steps:
S1.1: N is separately added into original wind-powered electricity generation time series x (t)GSecondary mean value is zero, and amplitude standard deviation is the Gauss white noise of constant
Sound ωi(t), new wind-powered electricity generation time series x is obtainedi(t):
xi(t)=x (t)+ωi(t), i=1,2 ..., NG
Wherein, NGFor the integer greater than 2;
S1.2: to each signal xi(t) Empirical mode decomposition decomposition is carried out respectively, obtains m intrinsic mode functions component and one
A residual components Ri(t):
Wherein, imfijIt (t) is that j-th of IMF that i-th addition white Gaussian noise is decomposed through Empirical mode decomposition divides
Amount;
S1.3: to all IMF component imfij(t) mean value computation is carried out, final IMF component IMF is obtainedj(t) and residual components
RES (t):
Wherein, IMFj(t) j-th of the IMF component decomposed for integrated Empirical mode decomposition;
S1.4: according to above-mentioned steps S1.1-S1.3, original wind power time series is divided by integrated Empirical mode decomposition
Solution is a series of IMF modal components.
3. the short-term wind power forecast method according to claim 1 based on hybrid algorithm, which is characterized in that described
Step S2 specifically includes the following steps:
S2.1: it decomposes: selected length of window L, whereinBy wind-powered electricity generation time series X=(x1,x2,…,xN), when N is wind-powered electricity generation
Between sequence samples amount, i.e., respectively S1 decomposes the multi-dimensional matrix that resulting each IMF component is converted into L × M:
Wherein, M=N-L+1;
S2.2: it calculates: calculating GGT, and carry out singular value decomposition and obtain L characteristic value: λ1≥λ2≥…≥λL>=0, each feature
Value represents corresponding singular value decomposition component and contributes the energy of original signal, and the corresponding feature vector of each characteristic value is respectively
U1,U2,…UL, and:
G=G1+G2+…Gd
Wherein, GiIndicate i-th of singular value decomposition (SVD) component,AndUiIt is i-th
The corresponding feature vector of characteristic value, d=max { i: λi> 0 };
S2.3: reconstruct: the singular value decomposition component of condition needed for selection meets constitutes a subset { GI,It can obtain
It arrives:
Wherein, λjFor j-th of characteristic value, UjFor the corresponding feature vector of j-th of characteristic value;
For GIContribution rate, and by the matrix G of L × MIIt is reconstructed into the main trend component Y=(y that length is N1,
y2,...,yN);
Enable L*=min (L, M), K*=max (L, M), if L≤M,It is on the contrary thenIt is calculate by the following formula weight
Structure sequence:
Wherein, gm,nRepresenting matrixIn m row n-th arrange element;
S2.4: according to step S2.1-S2.3, integrated Empirical mode decomposition is decomposed resulting in addition to first IMF1 component
Except each IMF component and the main trend component of RES component extract, and respectively correspond to obtain a remainder rk;
The remainder of each IMF component is superimposed the residual components R total as one to handle, it may be assumed that
Wherein, rkCorrespond to the remainder of k-th of IMF component.
4. the short-term wind power forecast method according to claim 2 based on hybrid algorithm, which is characterized in that described
How singular value decomposition component subset { G for reconstruct each IMF component main trend component is selected in step S2.3I, specifically
It operates as follows:
It is located at highest frequency range, contained noise energy since integrated Empirical mode decomposition decomposes resulting first IMF component
Amount at most, is handled so usually all being treated as high frequency noise, the noise energy in each IMF component of residue with order increasing
Add and constantly successively decrease, production decline law is as follows:
Wherein, EnkFor the energy of institute's Noise in k-th of IMF component, δ ≈ 0.719, ε ≈ 2.01;
The gross energy of each component can indicate are as follows:
Wherein, imfkIt (i) is i-th of element of k-th of IMF component;
By En1For the 1st IMF component IMF1The energy of middle institute's Noise, by IMF1All as high frequency noise, E can be obtainedn1=
E1, energy shared by signal in remaining IMF component (k >=2) are as follows:
Exk=Ek-Enk
So IMFkRatio shared by signal energy can calculate in (k >=2) are as follows:
Rk=Exk/Ek
On this basis, selection reconstruct Shi Neng represents input signal IMFkThe singular value decomposition component of (k >=2), specific method is such as
Under:
Work as Rk>=0.5, illustrate that signal is main component in the IMF component, by Exk=Ek-EnkCalculate H:
At this point, can represent this IMF sequence singular value decomposition component constitute subset as
Work as Rk≤ 0.5, illustrate that noise is main component in the IMF component, by Rk=Exk/EkCalculate H:
At this point, can represent this IMF sequence singular value decomposition component constitute subset as
5. the short-term wind power forecast method according to claim 1 based on hybrid algorithm, which is characterized in that described
Step S3 specifically includes the following steps:
S3.1: choosing the parameter of WAVELET PACKET DECOMPOSITION, sets three layers for Decomposition order, using wavelet packet decomposition algorithm by step 2
Obtained in the biggish submodule state of SE value decomposed, obtain one three layers of wavelet packet tree, form is binary tree;
S3.2: wavelet packet tree bottom node, i.e., the signal of eight nodes, so that it may the new son after obtaining double decomposition are read respectively
Mode sequence;
S3.3: IMF is respectively obtained1The two groups of new submodule state sequences of component and residual components R after the decomposition of WAVELET PACKET DECOMPOSITION technology
Column.
6. the short-term wind power forecast method according to claim 1 based on hybrid algorithm, which is characterized in that described
Step S4 specifically includes the following steps:
S4.1: initial phase utilizes initial training collectionInitial robust extreme learning machine model is established,
Constraint condition are as follows:
β0Weight, e are initially exported for hidden layer0For the training error of initial training collection, N0For initial training collection sample size, C is positive
Then change coefficient;H0For initial hidden layer output matrix, H0It can calculate are as follows:
Wherein, Q is node in hidden layer;
The constraint condition optimization problem can be iterated calculating by augmentation Lagrangian, its calculation formula is:
Wherein, μ=2N0/||Y0||1For penalty factor, λkFor the Lagrangian in kth time iteration;
According toAvailable βk+1And ek+1Final expression formula are as follows:
Wherein, I is unit matrix;
S4.2: the on-line study stage updates meter using recurrent least square method when+1 sample of kth of training set D arrives
It calculates hidden layer and exports weight, calculation formula are as follows:
Wherein, hk+1=[g (w1·xi+b1) … g(wQ·xi+bQ)];Error is predicted for new samples;ζk+1And εk+1Supplemented by
Help parameter, wherein work as ζk+1When=0, then Pk+1=Pk;
S4.3: initialization forgetting factor θ is 0≤θ0≤ 1, and update is iterated to θ according to the following formula:
νk+1=θk(νk+1)
Wherein, νk+1, ηk+1, γk+1It is auxiliary parameter, ρ is a positive number, and the value range of the initial value of ν, γ is [0,1].
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810997267.0A CN109376897A (en) | 2018-08-29 | 2018-08-29 | A kind of short-term wind power forecast method based on hybrid algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810997267.0A CN109376897A (en) | 2018-08-29 | 2018-08-29 | A kind of short-term wind power forecast method based on hybrid algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109376897A true CN109376897A (en) | 2019-02-22 |
Family
ID=65404281
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810997267.0A Pending CN109376897A (en) | 2018-08-29 | 2018-08-29 | A kind of short-term wind power forecast method based on hybrid algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109376897A (en) |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109993368A (en) * | 2019-04-08 | 2019-07-09 | 福州大学 | Power forecasting method based on unusual spectral factorization and shot and long term memory network |
CN110263915A (en) * | 2019-05-31 | 2019-09-20 | 广东工业大学 | A kind of wind power forecasting method based on deepness belief network |
CN110648017A (en) * | 2019-08-30 | 2020-01-03 | 广东工业大学 | Short-term impact load prediction method based on two-layer decomposition technology |
CN111353640A (en) * | 2020-02-26 | 2020-06-30 | 西南交通大学 | Method for constructing wind speed prediction model by combination method |
CN112034253A (en) * | 2020-09-21 | 2020-12-04 | 国网福建省电力有限公司 | MOA online monitoring method |
CN112034252A (en) * | 2020-09-21 | 2020-12-04 | 国网福建省电力有限公司 | MOA resistive current extraction method |
CN112070303A (en) * | 2020-09-08 | 2020-12-11 | 合肥工业大学 | Parameter-adaptive photovoltaic power ramp event hierarchical probabilistic prediction method |
CN112465225A (en) * | 2020-11-27 | 2021-03-09 | 云南电网有限责任公司电力科学研究院 | Wind power prediction method based on secondary modal decomposition and cascade deep learning |
CN112464923A (en) * | 2021-02-03 | 2021-03-09 | 四川轻化工大学 | Magnetic shoe internal defect detection method based on improved variational modal decomposition |
CN112712202A (en) * | 2020-12-29 | 2021-04-27 | 广东电网有限责任公司 | Short-term wind power prediction method and device, electronic equipment and storage medium |
CN113112011A (en) * | 2020-01-13 | 2021-07-13 | 中移物联网有限公司 | Data prediction method and device |
CN113723752A (en) * | 2021-07-26 | 2021-11-30 | 国网江苏省电力有限公司江阴市供电分公司 | Decomposition algorithm performance evaluation method in combined wind power prediction model |
CN113779861A (en) * | 2021-07-23 | 2021-12-10 | 国网河北省电力有限公司电力科学研究院 | Photovoltaic power prediction method and terminal equipment |
CN113836801A (en) * | 2021-09-13 | 2021-12-24 | 湖北工业大学 | Prediction method based on CEEMD and improved SSA-LSSVM |
CN113837465A (en) * | 2021-09-18 | 2021-12-24 | 湘潭大学 | Multi-stage campus power short-term load prediction method |
CN114035021A (en) * | 2021-10-08 | 2022-02-11 | 北京航空航天大学 | Circuit fault prediction method based on EEMD-Prophet |
CN114282440A (en) * | 2021-12-27 | 2022-04-05 | 淮阴工学院 | Robust identification method for adjusting system of pumped storage unit |
CN117635245A (en) * | 2023-11-30 | 2024-03-01 | 广东电力交易中心有限责任公司 | Power price prediction method and system based on multilevel frequency domain decomposition and IBM H optimization DELM |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106203723A (en) * | 2016-07-19 | 2016-12-07 | 河海大学 | Wind power short-term interval prediction method based on RT reconstruct EEMD RVM built-up pattern |
-
2018
- 2018-08-29 CN CN201810997267.0A patent/CN109376897A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106203723A (en) * | 2016-07-19 | 2016-12-07 | 河海大学 | Wind power short-term interval prediction method based on RT reconstruct EEMD RVM built-up pattern |
Non-Patent Citations (4)
Title |
---|
HAO YIN等: "An effective secondary decomposition approach for wind power forecasting using extreme learning machine trained by crisscross optimization", 《ENERGY CONVERSION AND MANAGEMENT》 * |
HUI LIU等: "Smart multi-step deep learning model for wind speed forecasting based on variational mode decomposition,singular spectrum analysis,LSTM network and ELM", 《ENERGY CONVERSION AND MANAGEMENT》 * |
XIANGANG PENG等: "A novel probabilistic wind speed forecasting based on combination of the adaptive ensemble of on-line sequential ORELM (Outlier Robust Extreme Learning Machine) and TVMCF (time-varying mixture copula function)", 《ENERGY CONVERSION AND MANAGEMENT》 * |
肖小兵等: "基于奇异谱分析的经验模态分解去噪方法", 《计算机工程与科学》 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109993368A (en) * | 2019-04-08 | 2019-07-09 | 福州大学 | Power forecasting method based on unusual spectral factorization and shot and long term memory network |
CN110263915A (en) * | 2019-05-31 | 2019-09-20 | 广东工业大学 | A kind of wind power forecasting method based on deepness belief network |
CN110648017A (en) * | 2019-08-30 | 2020-01-03 | 广东工业大学 | Short-term impact load prediction method based on two-layer decomposition technology |
CN113112011A (en) * | 2020-01-13 | 2021-07-13 | 中移物联网有限公司 | Data prediction method and device |
CN113112011B (en) * | 2020-01-13 | 2024-02-27 | 中移物联网有限公司 | Data prediction method and device |
CN111353640A (en) * | 2020-02-26 | 2020-06-30 | 西南交通大学 | Method for constructing wind speed prediction model by combination method |
CN111353640B (en) * | 2020-02-26 | 2022-04-08 | 西南交通大学 | Method for constructing wind speed prediction model by combination method |
CN112070303A (en) * | 2020-09-08 | 2020-12-11 | 合肥工业大学 | Parameter-adaptive photovoltaic power ramp event hierarchical probabilistic prediction method |
CN112070303B (en) * | 2020-09-08 | 2022-09-20 | 合肥工业大学 | Parameter-adaptive photovoltaic power ramp event hierarchical probabilistic prediction method |
CN112034252A (en) * | 2020-09-21 | 2020-12-04 | 国网福建省电力有限公司 | MOA resistive current extraction method |
CN112034253A (en) * | 2020-09-21 | 2020-12-04 | 国网福建省电力有限公司 | MOA online monitoring method |
CN112034253B (en) * | 2020-09-21 | 2022-06-07 | 国网福建省电力有限公司 | MOA online monitoring method |
CN112465225A (en) * | 2020-11-27 | 2021-03-09 | 云南电网有限责任公司电力科学研究院 | Wind power prediction method based on secondary modal decomposition and cascade deep learning |
CN112712202A (en) * | 2020-12-29 | 2021-04-27 | 广东电网有限责任公司 | Short-term wind power prediction method and device, electronic equipment and storage medium |
CN112464923B (en) * | 2021-02-03 | 2021-04-13 | 四川轻化工大学 | Magnetic shoe internal defect detection method based on improved variational modal decomposition |
CN112464923A (en) * | 2021-02-03 | 2021-03-09 | 四川轻化工大学 | Magnetic shoe internal defect detection method based on improved variational modal decomposition |
CN113779861A (en) * | 2021-07-23 | 2021-12-10 | 国网河北省电力有限公司电力科学研究院 | Photovoltaic power prediction method and terminal equipment |
CN113779861B (en) * | 2021-07-23 | 2023-08-22 | 国网河北省电力有限公司电力科学研究院 | Photovoltaic Power Prediction Method and Terminal Equipment |
CN113723752A (en) * | 2021-07-26 | 2021-11-30 | 国网江苏省电力有限公司江阴市供电分公司 | Decomposition algorithm performance evaluation method in combined wind power prediction model |
CN113836801A (en) * | 2021-09-13 | 2021-12-24 | 湖北工业大学 | Prediction method based on CEEMD and improved SSA-LSSVM |
CN113837465A (en) * | 2021-09-18 | 2021-12-24 | 湘潭大学 | Multi-stage campus power short-term load prediction method |
CN114035021A (en) * | 2021-10-08 | 2022-02-11 | 北京航空航天大学 | Circuit fault prediction method based on EEMD-Prophet |
CN114282440A (en) * | 2021-12-27 | 2022-04-05 | 淮阴工学院 | Robust identification method for adjusting system of pumped storage unit |
CN114282440B (en) * | 2021-12-27 | 2023-08-25 | 淮阴工学院 | Robust identification method for adjusting system of pumped storage unit |
CN117635245A (en) * | 2023-11-30 | 2024-03-01 | 广东电力交易中心有限责任公司 | Power price prediction method and system based on multilevel frequency domain decomposition and IBM H optimization DELM |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109376897A (en) | A kind of short-term wind power forecast method based on hybrid algorithm | |
Wang et al. | Bi-directional long short-term memory method based on attention mechanism and rolling update for short-term load forecasting | |
CN107220764A (en) | A kind of electricity sales amount Forecasting Methodology compensated based on preamble analysis and factor and device | |
CN108197773A (en) | Methods of electric load forecasting, load forecast device and terminal device | |
CN109886464B (en) | Low-information-loss short-term wind speed prediction method based on optimized singular value decomposition generated feature set | |
CN112364975A (en) | Terminal operation state prediction method and system based on graph neural network | |
CN109146186A (en) | A kind of short-term wind power forecast method based on double decomposition | |
Araya et al. | A Multi-Scale Model based on the Long Short-Term Memory for day ahead hourly wind speed forecasting | |
Lv et al. | EGA-STLF: A hybrid short-term load forecasting model | |
CN109840633A (en) | Photovoltaic output power predicting method, system and storage medium | |
Naduvil-Vadukootu et al. | Evaluating preprocessing strategies for time series prediction using deep learning architectures | |
CN114498619A (en) | Wind power prediction method and device | |
Khan et al. | A new hybrid approach of clustering based probabilistic decision tree to forecast wind power on large scales | |
Netsanet et al. | Input parameters selection and accuracy enhancement techniques in PV forecasting using Artificial Neural Network | |
CN115169742A (en) | Short-term wind power generation power prediction method | |
CN115034473A (en) | Electricity price prediction method, system and device | |
Tyass et al. | Wind speed prediction based on statistical and deep learning models | |
Sang et al. | Ensembles of gradient boosting recurrent neural network for time series data prediction | |
CN112766537B (en) | Short-term electric load prediction method | |
CN117371573A (en) | Time sequence prediction method, device and medium based on TrAdaBoost-LSTM | |
CN113837434A (en) | Solar photovoltaic power generation prediction method and device, electronic equipment and storage medium | |
CN117455536A (en) | Short-term coal price prediction method and system based on error compensation | |
CN116777039A (en) | Double-layer neural network wind speed prediction method based on training set segmentation and error correction | |
Qin et al. | Forecasting of China consumer price index based on EEMD and SVR method | |
Raphel | Artificial intelligence‐based wind forecasting using variational mode decomposition. |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190222 |
|
RJ01 | Rejection of invention patent application after publication |