CN109299430A - The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages - Google Patents
The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages Download PDFInfo
- Publication number
- CN109299430A CN109299430A CN201811165349.5A CN201811165349A CN109299430A CN 109299430 A CN109299430 A CN 109299430A CN 201811165349 A CN201811165349 A CN 201811165349A CN 109299430 A CN109299430 A CN 109299430A
- Authority
- CN
- China
- Prior art keywords
- wind speed
- decomposed
- time series
- learning machine
- stages
- 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
- 238000013277 forecasting method Methods 0.000 title claims abstract description 12
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 33
- 238000000034 method Methods 0.000 claims abstract description 24
- 230000008569 process Effects 0.000 claims abstract description 6
- 230000009466 transformation Effects 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000005311 autocorrelation function Methods 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 230000003416 augmentation Effects 0.000 claims description 2
- 230000001186 cumulative effect Effects 0.000 abstract description 2
- 238000013473 artificial intelligence Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000000714 time series forecasting Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 235000000332 black box Nutrition 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000004088 simulation 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Pure & Applied Mathematics (AREA)
- Strategic Management (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Operations Research (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Databases & Information Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Algebra (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to wind speed short-term forecast technical fields, disclose a kind of short-term wind speed forecasting method decomposed based on two stages with extreme learning machine, wherein two stages decomposition method is mainly used for decomposing original wind speed series, in decomposable process, original wind series are resolved into a series of intrinsic modal components with different frequency first with CEEMDAN, then second decomposition is carried out to reduce influence of the CEEMDAN high fdrequency component to prediction result using VMD intrinsic modal components highest to frequency, obtain a series of variation mode, finally each subsequence that two stages decompose is predicted using extreme learning machine model, and cumulative reconstruct is carried out to prediction result, obtain the predicted value of original wind series.The present invention can the random fluctuation tentatively to wind speed time series handle, reduce wind speed time series strong nonlinearity and the non-stationary influence to prediction result, further increase the precision of prediction of extreme learning machine prediction model.
Description
Technical field
It is the present invention relates to wind speed short-term forecast technical field, in particular to a kind of based on two stages decomposition and extreme learning machine
Short-term wind speed forecasting method.
Background technique
Existing wind speed forecasting method can be mainly divided into physical model, time series models and artificial intelligence model.Wherein
Physical model is primarily referred to as numerical weather forecast (NWP) model.The building of NWP model need to obtain from weather station or satellite including
The meteorological datas such as temperature, landform humidity, air pressure, wind speed and direction, this becomes difficult the building of NWP model and time-consuming.This
Outside, it needs to solve complicated mathematical model since NWP model is elongated, increases the time cost of NWP modeling.
Time series models and artificial intelligence model can be output and input according to history between mapping relations obtain wind
The predicted value of speed, modeling process is simpler, and data requirements amount is smaller.Time series models mainly include autoregressive moving average
(ARMA) model, autoregression integrate sliding average (ARIMA) model and ARIMA (ARIMAX) model with external item etc..
Time series models usually require that wind speed time series is stable and Normal Distribution, however wind speed time series generally has
There is the characteristics of non-linear and non-stationary, therefore, the ability predicted with time series models wind speed time series is limited.
Artificial intelligence model is a quasi-nonlinear black-box model, mainly includes artificial neural network (ANN), model, support
Vector machine (SVM) model, Adaptive Neuro-fuzzy Inference (ANFIS) model etc..ANN model mainly includes backpropagation mind
Through network (BPNN) model, radial basis function neural network (RBF) model, generalized regression nerve networks (GRNN) model, Elman
Neural network model and extreme learning machine (ELM) model of multilayer feedforward neural network (MLF) model etc..Wherein ELM model with
Its training speed is fast, Generalization Capability is good and it is famous, be widely used in forecasting wind speed in recent years.
Single artificial intelligence model achieves great successes, but single machine learning model in short-term wind speed forecasting field
Still it is not enough to capture the nonlinear characteristic of short-term wind speed time series, there are still very big rooms for improvement, in single artificial intelligence
On the basis of model, it is necessary to develop the mixing wind speed forecasting method of more discipline intersection applications.
Summary of the invention
Goal of the invention: aiming at the problems existing in the prior art, the present invention provides a kind of based on two stages decomposition and the limit
The short-term wind speed forecasting method of learning machine, can the random fluctuation tentatively to wind speed time series handle, reduce wind speed when
Between sequence strong nonlinearity and the non-stationary influence to prediction result, further increase the prediction of extreme learning machine prediction model
Precision.
Technical solution: the present invention provides a kind of short-term wind speed forecasting sides decomposed based on two stages with extreme learning machine
Method, which comprises the following steps: step 1: acquisition wind field history measured data establishes wind speed time series X (t), root
Training sample and test samples are classified as according to the data cases of the wind speed time series X (t)；Step 2: using CEEMDAN points
The wind speed time series X (t) is decomposed into the intrinsic modal components and a residual sequence of several different frequencies by solution；
CEEMDAN is as follows to the decomposable process of the wind speed time series X (t): (1) on the basis of the wind speed time series X (t)
It generates one group and adds sequence of making an uproar:
Xi(t)=X (t)+p0ωi(t) (1)
In formula, Xi(t) indicate plus make an uproar sequence, and X (t) is wind speed time series；ωi(t) one group of (i=1 ..., I) expression is equal
Value is 0, the white Gaussian noise that variance is 1, and wherein I indicates to realize number, that is, is added to the sequence number of white noise, 1 time EMD is decomposed
Referred to as 1 time realization；p0It is the noise coefficient for controlling additional noise and original signal signal-to-noise ratio；(2) it is decomposed using EMD to each
Xi(t) it is decomposed, obtains the intrinsic modal components IMF of first time EMD decomposition1 i(t), then the first rank eigen mode of CEEMDAN
State component are as follows:
(3) the first rank residue signal that CEEMDAN is decomposed is calculatedAnd first rank that introduces plus make an uproar
The signal r to be decomposed of component and residue signal composition newly1(t)+p1E1(wi(t)), continue with EMD decompose to obtain first it is intrinsic
Modal components calculate the intrinsic modal components of second-order of CEEMDAN:
In formula, E1() indicates the function for asking EMD to decompose first intrinsic modal components, p1Indicate noise coefficient；(4) weight
Multiple step (3) are decomposed when the extreme point number of residual signals is no more than two and are completed；Step 3: variation mode decomposition is used
CEEMDAN is decomposed the highest intrinsic modal components of resulting frequency and carries out second decomposition by VMD, is obtained a series of in difference
The variation mode u of frequency of heartk；The number K that variation mode is determined using centre frequency observation enables K=1,2 ..., k, uses
VMD pairsThe highest intrinsic modal components of frequencySignal is decomposed, if as K=k, the center frequency for the variation mode that VMD is decomposed
Rate is more close, then K=k-1 is the variation mode number finally chosen；Suitable τ is chosen using residual error evaluation index REI；Step
Rapid four: determining that CEEMDAN and VMD two stages are decomposed resulting all subsequence prediction models according to partial autocorrelation function PACF
Input variable, and determine output variable；Step 5: the input that step 4 is determined and output variable normalize, and utilize
Training sample trains extreme learning machine model；Step 6: the test samples are inputted into trained extreme learning machine model simultaneously
Obtained predicted value renormalization is obtained into the predicted value of each subsequence, when summing to obtain wind speed to the predicted value of all subsequences
Between sequence predicted value；Step 7: root-mean-square error RMSE, mean absolute error MAE and average absolute measure error are used
The performance of tri- evaluation index evaluation wind speed time series predicting models of MASE.
Further, specific step is as follows for the step 3: it is limited that One-dimension Time Series x (t) is resolved into K by VMD
The variation mode u of bandwidthk(k=1,2 ..., K), it is assumed that each variation mode ukCentre frequency be ωk, in order to determine VMD points
Each group of u of solutionkAnd ωk, need to solve following constrained optimization problem:
In formula, { uk} :={ u1..., uKIndicate variation mode set, { ωk} :={ ω1..., ωKIndicate variation mould
The corresponding centre frequency of state,Indicate that derivation symbol, δ (t) indicate Dirac distribution, j indicates imaginary unit, and * indicates convolution fortune
It calculates；It introduces secondary penalty factor α and Lagrange multiplier λ and solves above-mentioned constrained optimization problem, then the augmentation of former optimization problem
Lagrange function can be described as:
Unconstrained Optimization Problem in formula (5) can be solved by alternating direction multipliers method, obtain different variation moulds
State and corresponding centre frequency, wherein the variation mode u in time domainkIt can be converted by Fourier's equilong transformation to frequency domain:
In formula, ω is frequency,For the Fourier transformation of signal x (ω),For the Fourier transformation of λ (ω),It indicatesFourier transformation, iff () indicates to seek the function of inverse Fourier transform,Analytic signal is sought in expression
Real part；Centre frequency ωkUpdate mode it is as follows:
In formula,Indicate ukThe Fourier transformation of (ω)；VMD is obtained by ceaselessly iterating to calculate formula (6) and (8)
Shown in the output of decomposition such as formula (7), wherein the Lagrange multiplier in nth iteration updates according to the following formula:
In formula, τ is iteration coefficient.
Further, in the step 3, the REI is calculate by the following formula:
In formula, i is the serial number of sample point, and N is the total number of sample point, and x (i) is original time series, ukIt (i) is original
K-th of the variation mode that time series VMD is decomposed.
Preferably, τ=0 is enabled: 0.01: 1, the REI under different τ values is calculated, the selection the smallest τ value of REI is optimal value.Wind speed
Time series forecasting generally requires REI index minimum, enables τ=0: 0.01: 1, the REI under different τ values is calculated, it is minimum to choose REI
τ value be optimal value.
Preferably, in the step 4, for each subsequence, the PACF value under different time delay, selection are calculated
Historical series of the PACF value except 95% confidence level are 48 as prediction input variable, maximum time delay.
Preferably, in the step 5, the hidden layer node number of extreme learning machine model uses grid search (GS)
Algorithm determines that the search range of GS algorithm is set as [m, 2n+20], step-size in search 1, and n indicates input layer node number, if n >
10, m=2n-20, else n=1.
Preferably, in the step 7, RMSE, MAE and MASE are defined as follows:
In formula, qp(i) and qo(i) predicted value and measured value of i-th of sample of wind speed time series are respectively indicated, N indicates sample
This number.
The utility model has the advantages that the present invention is for the non-linear and non-stationary of short-term wind speed time series, the invention proposes one
Kind is decomposed based on two stages and the short-term wind speed forecasting model of extreme learning machine, and wherein two stages decomposition method is mainly used for original
Wind speed series is decomposed, in decomposable process, first with CEEMDAN by original wind series resolve into it is a series of have not
Then the intrinsic modal components of same frequency carry out second decomposition using VMD intrinsic modal components highest to frequency to reduce
Influence of the CEEMDAN high fdrequency component to prediction result obtains a series of variation mode submodule states, finally uses extreme learning machine mould
Type predicts each subsequence that two stages decompose, and carries out cumulative reconstruct to prediction result, obtains original wind speed sequence
The predicted value of column.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, have below beneficial to effect
Fruit:
1) present invention carries out first stage decomposition to wind speed time series using CEEMDAN first, by wind speed time series
Be decomposed into several relatively stable variation mode, can the random fluctuation tentatively to wind speed time series handle, reduce
The strong nonlinearity of wind speed time series and the non-stationary influence to prediction result；
2) interference that the high fdrequency component generated generates prediction result is decomposed for first stage CEEMDAN, the present invention adopts
Second decomposition is carried out to CEEMDAN high fdrequency component with VMD decomposition, gained is then decomposed to the two-stage using extreme learning machine model
All variation mode carry out prediction summation, the precision of prediction of extreme learning machine prediction model can be further increased.
Detailed description of the invention
Fig. 1 is the short-term wind speed forecasting model flow figure provided by the invention decomposed based on two stages with extreme learning machine；
Fig. 2 is wind speed time series two stages decomposition result figure provided by the invention；
Fig. 3 is wind speed time series Single-step Prediction result figure provided by the invention.
Specific embodiment
The present invention is described in detail with reference to the accompanying drawing.
Embodiment 1:
The air speed data that the present invention chooses Inner Mongol wind power plant is embodiment, Case Simulation is carried out, to verify the present invention
Effect.Fig. 1 is the short-term wind speed forecasting method flow chart provided by the invention decomposed based on two stages with extreme learning machine；This
Prediction technique the following steps are included:
Step 1: total 672 observation points of continuous one week air speed data Inner Mongol wind power plant in March, 2016 are chosen and are made
For sample data, wind speed time series is constituted, is divided into 15 minutes between wind-speed sample, chooses preceding the 520 of entire wind speed time series
A observation point is as training sample, and remaining 152 observation points are as test samples.
Step 2: empirical mode decomposition CEEMDAN is completely integrate using adaptive noise and is decomposed wind speed time series point
Solution is the intrinsic modal components and a residual sequence of several different frequencies, change of the wind speed time series after CEEMDAN is decomposed
Divide shown in mode such as Fig. 2 (a).
CEEMDAN is that a kind of processing to grow up on the basis of Conventional wisdom mode decomposition EMD is non-linear and non-flat
The data analysing method of steady signal, the main thought of EMD method are the different Fluctuation Scales by signal by complicated original signal
Resolve into limited intrinsic modal components.Based on EMD method, the data analysing method of aid in noise is repeatedly added into original signal
Referred to as gather empirical mode decomposition EEMD, EEMD can effectively improve the drawbacks of EMD mode mixes.However, by limited times iteration
Afterwards, reconstruction error caused by white noise added by EEMD is not completely counterbalanced by, so that the precision of reconstruction time sequence is influenced, into
One step influences precision of prediction.Although reconstructed error can be reduced to a certain extent by increasing the number of iterations, cost is calculated
Also increase therewith, and CEEMDAN decomposition can save while reducing reconstruction error and calculate cost.
CEEMDAN is as follows to the decomposable process of wind speed time series X (t):
(1) one group is generated on the basis of wind speed time series X (t) add sequence of making an uproar:
Xi(t)=X (t)+p0ωi(t) (1)
In formula, Xi(t) indicate plus make an uproar sequence, and X (t) is wind speed time series；ωi(t) one group of (i=1 ..., I) expression is equal
Value is 0, the white Gaussian noise that variance is 1, and wherein I indicates to realize number, that is, is added to the sequence number of white noise, 1 time EMD is decomposed
Referred to as 1 time realization；p0It is the noise coefficient for controlling additional noise and original signal signal-to-noise ratio.
(2) it is decomposed using EMD to each Xi(t) it is decomposed, obtains the intrinsic modal components of first time EMD decomposition
IMF1 i(t).The then intrinsic modal components of the first rank of CEEMDAN are as follows:
(3) the first rank residue signal that CEEMDAN is decomposed is calculatedAnd first rank that introduces plus make an uproar
The signal r to be decomposed of component and the first rank residue signal composition newly1(t)+p1E1(wi(t)), continue to be decomposed to obtain first with EMD
A intrinsic modal components calculate the intrinsic modal components of second-order of CEEMDAN:
In formula, E1() indicates the function for asking EMD to decompose first intrinsic modal components, p1Indicate noise coefficient.
(4) step (3) are repeated and decompose completion when the extreme point number of residual signals is no more than two.
Step 3: CEEMDAN is decomposed by the highest component of resulting frequency using variation mode decomposition VMD and carries out secondary point
Solution, obtains a series of variation mode u with different center frequencyk, the variation mode such as Fig. 2 of high fdrequency component after VMD is decomposed
(b) shown in.
VMD is a kind of common Non-stationary Signal Analysis method, and it is a limited that One-dimension Time Series X (t) can be resolved into K
The variation mode or variation mode u of bandwidthk(k=1,2 ..., K), it is assumed that each variation mode ukCentre frequency be ωk, it is
Each group of u that determining VMD is decomposedkAnd ωk, need to solve following constrained optimization problem:
In formula, { u is indicatedk} :={ u1..., uKIndicate variation mode set, { ωk} :={ ω1..., ωKIndicate to become
Divide the corresponding centre frequency of mode,Indicate that derivation symbol, δ (t) indicate Dirac distribution, j indicates imaginary unit, and * indicates volume
Product operation.
It introduces secondary penalty factor α and Lagrange multiplier λ and solves above-mentioned constrained optimization problem, then the increasing of former optimization problem
Wide Lagrange function can be described as:
Unconstrained Optimization Problem in formula (2) can be solved by alternating direction multipliers method, obtain different variation moulds
State and corresponding centre frequency, wherein the mode u in time domainkIt can be converted by Fourier's equilong transformation to frequency domain:
In formula, ω is frequency,For the Fourier transformation of signal x (ω),For the Fourier transformation of λ (ω),It indicatesFourier transformation, iff () indicates to seek the function of inverse Fourier transform,Analytic signal is sought in expression
Real part.
Centre frequency ωkUpdate mode it is as follows:
In formula,Indicate ukThe Fourier transformation of (ω).
It is obtained shown in the output such as formula (7) of VMD decomposition by ceaselessly iterating to calculate formula (6) and (8), wherein n-th changes
Lagrange multiplier in generation updates according to the following formula:
In formula, τ is iteration coefficient.
VMD decomposes the number K for needing to determine variation mode in advance and iteration coefficient τ, choosing of the performance that VMD is decomposed to K and τ
Select quite sensitive, it is therefore desirable to determine suitable K and τ to carry out the prediction of next stage.The present invention is seen using centre frequency
The method of examining determines parameter K, enables K=1,2 ..., k, is decomposed using VMD intrinsic mode component signal highest to frequency, if working as
When K=k, the centre frequency of VMD variation mode is more close, then K=k-1 is the variation mode number finally chosen；Using residual
Poor evaluation index REI chooses suitable τ, and REI is calculate by the following formula:
In formula, i is the serial number of sample point, and N is the total number of sample point, and x (i) is original time series, ukIt (i) is original
K-th of the variation mode that time series VMD is decomposed.
Wind speed time series forecasting generally requires REI index minimum, enables τ=0: 0.01: 1, the REI under different τ values is calculated,
The selection the smallest τ value of REI is optimal value.
Step 4: determine that CEEMDAN and VMD two stages are decomposed resulting all sons according to partial autocorrelation function (PACF)
The input variable of sequential forecasting models.For each variation mode, the PACF value under different time delay is calculated, PACF value is selected
For historical series except 95% confidence level as prediction input variable, maximum time delay is 48, and determines output variable.
Step 5: variable normalization is output and input by what step 4 determined, and is learnt using the training sample training limit
Machine model, wherein the hidden layer node number of extreme learning machine model is determined using grid search (GS) algorithm, and GS algorithm is searched
Rope range is set as [m, 2n+20], step-size in search 1, and n indicates input layer node number, if n > 10, m=2n-20, else n
=1.
Step 6: by test samples input trained extreme learning machine model and by predicted value renormalization obtain
To the predicted value of each subsequence, the predicted value of all subsequences is summed to obtain the predicted value of wind speed time series.
Step 7: root-mean-square error RMSE, mean absolute error MAE and average absolute measure error MASE tri- is used to comment
The performance of valence metrics evaluation prediction model, RMSE, MAE and MASE are defined as follows:
In formula, qp(i) and qo(i) predicted value and measured value of i-th of sample of wind speed time series are respectively indicated, N indicates sample
This number.
Table 1 is single extreme learning machine (ELM) model provided by the invention, the single decomposition model based on CEEMDAN
The probative term single step of CEEMDAN-ELM (abbreviation CELM) and two stages decomposition model CEEMDAN-VMD-ELM (abbreviation CVELM) and
Multi-step prediction result indicator-specific statistics table.
Table 1
As shown in Table 1,3 kinds of wind speed time series predicting models can preferably carry out 15min wind speed time series pre-
It surveys, by comparing RMSE, MAE and the MASE of this 3 kinds of models, it can be found that the performance of CVELM model is superior to other 2 moulds
Type illustrates that the prediction effect of CVELM model is best.
It is also known by table 1, RMSE, MAE and the MASE of two stages decomposition model (CVELM) are than corresponding single decomposition model
(CELM) small, illustrate that two stages decomposition method can effectively decompose wind speed time series, reduces high fdrequency component non-stationary
Property influence to prediction result, and then improve model prediction accuracy.
Fig. 3 illustrates the predicted value and measured value comparison diagram of 3 kinds of wind speed time series predicting models.From the figure 3, it may be seen that this hair
Bright mentioned prediction model can preferably predict wind speed time series.
The technical concepts and features of above embodiment only to illustrate the invention, its object is to allow be familiar with technique
People cans understand the content of the present invention and implement it accordingly, and it is not intended to limit the scope of the present invention.It is all according to the present invention
The equivalent transformation or modification that Spirit Essence is done, should be covered by the protection scope of the present invention.
Claims (7)
1. a kind of short-term wind speed forecasting method decomposed based on two stages with extreme learning machine, which is characterized in that including following step
It is rapid:
Step 1: acquisition wind field history measured data establishes wind speed time series X (t), according to the wind speed time series X's (t)
Data cases are classified as training sample and test samples；
Step 2: the wind speed time series X (t) is divided using adaptive noise complete set empirical mode decomposition CEEMDAN
Solution is the intrinsic modal components of several different frequencies；
CEEMDAN is as follows to the decomposable process of the wind speed time series X (t):
(1) one group is generated on the basis of the wind speed time series X (t) add sequence of making an uproar:
Xi(t)=X (t)+p0ωi(t) (1)
In formula, Xi(t) indicate plus make an uproar sequence, and X (t) is wind speed time series；ωi(t) (i=1 ..., I) indicates that a class mean is
0, the white Gaussian noise that variance is 1, wherein I indicates to realize number, that is, is added to the sequence number of white noise, and 1 EMD decomposition is known as
1 realization；p0It is the noise coefficient for controlling additional noise and original signal signal-to-noise ratio；
(2) it is decomposed using EMD to each Xi(t) it is decomposed, obtains the intrinsic modal components IMF of first time EMD decomposition1 i
(t), then the intrinsic modal components of the first rank of CEEMDAN are as follows:
(3) the first rank residue signal that CEEMDAN is decomposed is calculatedAnd first order component that introduces plus make an uproar
The new signal r to be decomposed with the first rank residue signal composition1(t)+p1E1(wi(t)), continue to be decomposed to obtain first sheet with EMD
Modal components are levied, the intrinsic modal components of second-order of CEEMDAN are calculated:
In formula, E1() indicates the function for asking EMD to decompose first intrinsic modal components, p1Indicate noise coefficient；
(4) step (3) are repeated and decompose completion when the extreme point number of residual signals is no more than two；
Step 3: CEEMDAN is decomposed by the highest intrinsic modal components of resulting frequency using variation mode decomposition VMD and carries out two
Secondary decomposition obtains a series of variation mode u with different center frequencyk；
The number K that variation mode is determined using centre frequency observation enables K=1,2 ..., k, highest to frequency using VMD
Intrinsic modal components are decomposed, if the centre frequency for the variation mode that VMD is decomposed is more close, then K=k-1 as K=k
For the variation mode number finally chosen；
Suitable τ is chosen using residual error evaluation index REI；
Step 4: it is pre- to determine that CEEMDAN and VMD two stages are decomposed resulting all subsequences according to partial autocorrelation function PACF
The input variable of model is surveyed, and determines output variable；
Step 5: the input that step 4 is determined and output variable normalize, and utilize training sample training extreme learning machine
Model；
Step 6: the test samples are inputted into trained extreme learning machine model and obtain obtained predicted value renormalization
To the predicted value of each subsequence, the predicted value of all subsequences is summed to obtain the predicted value of wind speed time series；
Step 7: using root-mean-square error RMSE, and mean absolute error MAE and average tri- evaluations of absolute measure error MASE refer to
The performance of mark evaluation wind speed time series predicting model.
2. the short-term wind speed forecasting method according to claim 1 decomposed based on two stages with extreme learning machine, feature
Be, specific step is as follows for the step 3: One-dimension Time Series x (t) is resolved into the variation mould of K finite bandwidth by VMD
State uk(k=1,2 ..., K), it is assumed that each variation mode ukCentre frequency be ωk, in order to determine each group of u of VMD decompositionk
And ωk, need to solve following constrained optimization problem:
In formula, { uk} :={ u1..., uKIndicate variation mode set, { ωk} :={ ω1..., ωKIndicate variation mode pair
The centre frequency answered,Indicate that derivation symbol, δ (t) indicate Dirac distribution, j indicates imaginary unit, and * indicates convolution algorithm；
It introduces secondary penalty factor α and Lagrange multiplier λ and solves above-mentioned constrained optimization problem, then the augmentation of former optimization problem
Lagrange function can be described as:
Unconstrained Optimization Problem in formula (5) can be solved by alternating direction multipliers method, obtain different variation mode and
Corresponding centre frequency, wherein the variation mode u in time domainkIt can be converted by Fourier's equilong transformation to frequency domain:
In formula, ω is frequency,For the Fourier transformation of signal x (ω),For the Fourier transformation of λ (ω),Table
ShowFourier transformation, iff () indicates to seek the function of inverse Fourier transform,The real part of analytic signal is sought in expression；
Centre frequency ωkUpdate mode it is as follows:
In formula,Indicate ukThe Fourier transformation of (ω)；
It is obtained shown in the output such as formula (7) of VMD decomposition by ceaselessly iterating to calculate formula (6) and (8), wherein in nth iteration
Lagrange multiplier update according to the following formula:
In formula, τ is iteration coefficient.
3. the short-term wind speed forecasting method according to claim 2 decomposed based on two stages with extreme learning machine, feature
It is, in the step 3, the REI is calculate by the following formula:
In formula, i is the serial number of sample point, and N is the total number of sample point, and x (i) is original time series, ukIt (i) is original time
K-th of the variation mode that sequence VMD is decomposed.
4. the short-term wind speed forecasting method according to claim 3 decomposed based on two stages with extreme learning machine, feature
It is, enables τ=0: 0.01: 1, the REI under different τ values is calculated, the selection the smallest τ value of REI is optimal value.
5. the short-term wind speed forecasting according to any one of claim 1 to 4 decomposed based on two stages with extreme learning machine
Method, which is characterized in that in the step 4, for each subsequence, calculate the PACF value under different time delay, selection
Historical series of the PACF value except 95% confidence level are 48 as prediction input variable, maximum time delay.
6. the short-term wind speed forecasting according to any one of claim 1 to 4 decomposed based on two stages with extreme learning machine
Method, which is characterized in that in the step 5, the hidden layer node number of extreme learning machine model uses grid search (GS)
Algorithm determines that the search range of GS algorithm is set as [m, 2n+20], step-size in search 1, and n indicates input layer node number, if n >
10, m=2n-20, else n=1.
7. the short-term wind speed forecasting according to any one of claim 1 to 4 decomposed based on two stages with extreme learning machine
Method, which is characterized in that in the step 7, RMSE, MAE and MASE are defined as follows:
In formula, qp(i) and qo(i) predicted value and measured value of i-th of sample of wind speed time series are respectively indicated, N indicates sample point
Number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811165349.5A CN109299430A (en) | 2018-09-30 | 2018-09-30 | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811165349.5A CN109299430A (en) | 2018-09-30 | 2018-09-30 | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109299430A true CN109299430A (en) | 2019-02-01 |
Family
ID=65161702
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811165349.5A Pending CN109299430A (en) | 2018-09-30 | 2018-09-30 | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109299430A (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886354A (en) * | 2019-03-06 | 2019-06-14 | 国家卫星海洋应用中心 | Wind field fusion method and device |
| CN110046756A (en) * | 2019-04-08 | 2019-07-23 | 东南大学 | Short-time weather forecasting method based on Wavelet Denoising Method and Catboost |
| CN110082841A (en) * | 2019-04-18 | 2019-08-02 | 东华大学 | A kind of short-term wind speed forecasting method |
| CN110197306A (en) * | 2019-05-31 | 2019-09-03 | 东北大学 | A kind of dynamic glucose on-line prediction method based on VMD-PACF-BP model |
| CN110298097A (en) * | 2019-06-21 | 2019-10-01 | 中科国风科技有限公司 | A kind of fan blade of wind generating set Lay up design method |
| CN110363141A (en) * | 2019-07-15 | 2019-10-22 | 郑州大学 | Method for diagnosing gas pressure regulator, governor failure |
| CN110514294A (en) * | 2019-08-30 | 2019-11-29 | 鞍钢矿业爆破有限公司 | A kind of blasting vibration signal noise-reduction method based on EMD and VMD |
| CN110738010A (en) * | 2019-10-17 | 2020-01-31 | 湖南科技大学 | Wind power plant short-term wind speed prediction method integrated with deep learning model |
| CN111105321A (en) * | 2019-12-03 | 2020-05-05 | 湘潭大学 | Power load prediction method based on EEMD secondary decomposition |
| CN111178616A (en) * | 2019-12-24 | 2020-05-19 | 淮阴工学院 | Wind speed prediction method based on negative correlation learning and regularization extreme learning machine integration |
| CN111523728A (en) * | 2020-04-23 | 2020-08-11 | 东北电力大学 | Four-stage hybrid short-term wind direction prediction method |
| CN111553513A (en) * | 2020-04-14 | 2020-08-18 | 太原理工大学 | Medium-and-long-term runoff prediction method based on quadratic decomposition and echo state network |
| CN111738073A (en) * | 2020-05-16 | 2020-10-02 | 北京信息科技大学 | System and method for identifying optical fiber signals |
| CN111950759A (en) * | 2020-05-18 | 2020-11-17 | 武汉科技大学 | Short-term wind speed prediction method based on two-stage decomposition, LSTM and AT |
| CN112183814A (en) * | 2020-08-26 | 2021-01-05 | 东南大学 | Short-term wind speed prediction method |
| CN112232565A (en) * | 2020-10-15 | 2021-01-15 | 武汉科技大学 | Two-stage time sequence prediction method, prediction system, terminal and medium |
| CN112417762A (en) * | 2020-11-25 | 2021-02-26 | 中国民航大学 | Sector flow short-term prediction method based on decomposition integration methodology |
| CN112613650A (en) * | 2020-12-14 | 2021-04-06 | 燕山大学 | Novel three-stage short-term power load prediction method |
| CN113011669A (en) * | 2021-03-30 | 2021-06-22 | 北京科技大学 | Method for predicting monthly stock quantity of live pigs |
| CN113571136A (en) * | 2021-06-08 | 2021-10-29 | 广州地铁设计研究院股份有限公司 | Polycyclic aromatic hydrocarbon prediction method, device, medium and equipment |
| CN113642767A (en) * | 2021-07-09 | 2021-11-12 | 武汉科技大学 | Multi-dimensional feature combination prediction method based on MI-VMD-DA-EDLSTM-VEC |
| CN115438301A (en) * | 2022-10-26 | 2022-12-06 | 北京建筑大学 | Equipment operation trend prediction method based on ICEEMDAN quadratic decomposition coupling informar model |
| CN116128124A (en) * | 2023-01-09 | 2023-05-16 | 北京建筑大学 | Building energy consumption prediction method based on abnormal energy value processing and time sequence decomposition |
| CN116780776A (en) * | 2023-06-29 | 2023-09-19 | 淮阴工学院 | Chemical industry park photovoltaic monitoring system and method based on improved sparrow algorithm |
| CN117878933A (en) * | 2024-03-13 | 2024-04-12 | 南京信息工程大学 | Wind power prediction method based on abrupt weather environment recognition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106557840A (en) * | 2016-11-14 | 2017-04-05 | 中南大学 | A kind of high wind line of high-speed railway wind speed adaptive decomposition Forecasting Methodology |
| CN106897794A (en) * | 2017-01-12 | 2017-06-27 | 长沙理工大学 | A kind of wind speed forecasting method based on complete overall experience mode decomposition and extreme learning machine |
| CN108090621A (en) * | 2017-12-28 | 2018-05-29 | 华中科技大学 | A kind of short-term wind speed forecasting method and system based on global optimization stage by stage |
-
2018
- 2018-09-30 CN CN201811165349.5A patent/CN109299430A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106557840A (en) * | 2016-11-14 | 2017-04-05 | 中南大学 | A kind of high wind line of high-speed railway wind speed adaptive decomposition Forecasting Methodology |
| CN106897794A (en) * | 2017-01-12 | 2017-06-27 | 长沙理工大学 | A kind of wind speed forecasting method based on complete overall experience mode decomposition and extreme learning machine |
| CN108090621A (en) * | 2017-12-28 | 2018-05-29 | 华中科技大学 | A kind of short-term wind speed forecasting method and system based on global optimization stage by stage |
Non-Patent Citations (2)
| Title |
|---|
| WENYU ZHANG等: "A combined model based on CEEMDAN and modified flower pollination algorithm for wind speed forecasting", 《ENERGYCONVERSION & MANAGEMENT》 * |
| 殷豪 等: "基于可变模式分解和NWCSO优化极限学习机的短期风速预测", 《电力建设》 * |
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886354A (en) * | 2019-03-06 | 2019-06-14 | 国家卫星海洋应用中心 | Wind field fusion method and device |
| CN109886354B (en) * | 2019-03-06 | 2020-12-22 | 国家卫星海洋应用中心 | Wind field fusion method and device |
| CN110046756A (en) * | 2019-04-08 | 2019-07-23 | 东南大学 | Short-time weather forecasting method based on Wavelet Denoising Method and Catboost |
| CN110082841A (en) * | 2019-04-18 | 2019-08-02 | 东华大学 | A kind of short-term wind speed forecasting method |
| CN110197306A (en) * | 2019-05-31 | 2019-09-03 | 东北大学 | A kind of dynamic glucose on-line prediction method based on VMD-PACF-BP model |
| CN110298097A (en) * | 2019-06-21 | 2019-10-01 | 中科国风科技有限公司 | A kind of fan blade of wind generating set Lay up design method |
| CN110298097B (en) * | 2019-06-21 | 2023-07-18 | 中科国风科技有限公司 | Wind turbine blade layering design method of wind generating set |
| CN110363141B (en) * | 2019-07-15 | 2021-09-17 | 郑州大学 | Method for diagnosing a fault in a gas pressure regulator |
| CN110363141A (en) * | 2019-07-15 | 2019-10-22 | 郑州大学 | Method for diagnosing gas pressure regulator, governor failure |
| CN110514294A (en) * | 2019-08-30 | 2019-11-29 | 鞍钢矿业爆破有限公司 | A kind of blasting vibration signal noise-reduction method based on EMD and VMD |
| CN110738010A (en) * | 2019-10-17 | 2020-01-31 | 湖南科技大学 | Wind power plant short-term wind speed prediction method integrated with deep learning model |
| CN111105321A (en) * | 2019-12-03 | 2020-05-05 | 湘潭大学 | Power load prediction method based on EEMD secondary decomposition |
| CN111178616A (en) * | 2019-12-24 | 2020-05-19 | 淮阴工学院 | Wind speed prediction method based on negative correlation learning and regularization extreme learning machine integration |
| CN111178616B (en) * | 2019-12-24 | 2022-03-22 | 淮阴工学院 | Wind speed prediction method based on negative correlation learning and regularization extreme learning machine integration |
| CN111553513A (en) * | 2020-04-14 | 2020-08-18 | 太原理工大学 | Medium-and-long-term runoff prediction method based on quadratic decomposition and echo state network |
| CN111523728A (en) * | 2020-04-23 | 2020-08-11 | 东北电力大学 | Four-stage hybrid short-term wind direction prediction method |
| CN111523728B (en) * | 2020-04-23 | 2022-05-10 | 东北电力大学 | Four-stage hybrid short-term wind direction prediction method |
| CN111738073A (en) * | 2020-05-16 | 2020-10-02 | 北京信息科技大学 | System and method for identifying optical fiber signals |
| CN111950759A (en) * | 2020-05-18 | 2020-11-17 | 武汉科技大学 | Short-term wind speed prediction method based on two-stage decomposition, LSTM and AT |
| CN112183814A (en) * | 2020-08-26 | 2021-01-05 | 东南大学 | Short-term wind speed prediction method |
| CN112232565A (en) * | 2020-10-15 | 2021-01-15 | 武汉科技大学 | Two-stage time sequence prediction method, prediction system, terminal and medium |
| CN112417762B (en) * | 2020-11-25 | 2022-04-29 | 中国民航大学 | Sector flow short-term prediction method based on decomposition integration methodology |
| CN112417762A (en) * | 2020-11-25 | 2021-02-26 | 中国民航大学 | Sector flow short-term prediction method based on decomposition integration methodology |
| CN112613650A (en) * | 2020-12-14 | 2021-04-06 | 燕山大学 | Novel three-stage short-term power load prediction method |
| CN113011669A (en) * | 2021-03-30 | 2021-06-22 | 北京科技大学 | Method for predicting monthly stock quantity of live pigs |
| CN113011669B (en) * | 2021-03-30 | 2022-08-02 | 北京科技大学 | Method for predicting monthly stock quantity of live pigs |
| CN113571136A (en) * | 2021-06-08 | 2021-10-29 | 广州地铁设计研究院股份有限公司 | Polycyclic aromatic hydrocarbon prediction method, device, medium and equipment |
| CN113642767A (en) * | 2021-07-09 | 2021-11-12 | 武汉科技大学 | Multi-dimensional feature combination prediction method based on MI-VMD-DA-EDLSTM-VEC |
| CN115438301A (en) * | 2022-10-26 | 2022-12-06 | 北京建筑大学 | Equipment operation trend prediction method based on ICEEMDAN quadratic decomposition coupling informar model |
| CN115438301B (en) * | 2022-10-26 | 2023-06-20 | 北京建筑大学 | Equipment operation trend prediction method based on ICEEMDAN (information and communication technology) secondary decoupling index model |
| CN116128124A (en) * | 2023-01-09 | 2023-05-16 | 北京建筑大学 | Building energy consumption prediction method based on abnormal energy value processing and time sequence decomposition |
| CN116780776A (en) * | 2023-06-29 | 2023-09-19 | 淮阴工学院 | Chemical industry park photovoltaic monitoring system and method based on improved sparrow algorithm |
| CN117878933A (en) * | 2024-03-13 | 2024-04-12 | 南京信息工程大学 | Wind power prediction method based on abrupt weather environment recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299430A (en) | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages | |
| CN108256697B (en) | Prediction method for short-term load of power system | |
| Lee et al. | Short-term wind power ensemble prediction based on Gaussian processes and neural networks | |
| CN110309603B (en) | Short-term wind speed prediction method and system based on wind speed characteristics | |
| CN110991721A (en) | Short-term wind speed prediction method based on improved empirical mode decomposition and support vector machine | |
| CN110119086B (en) | Tomato greenhouse environmental parameter intelligent monitoring device based on ANFIS neural network | |
| CN108647807B (en) | River flow prediction method | |
| CN114970952B (en) | Photovoltaic output short-term prediction method and system considering environmental factors | |
| CN106649479A (en) | Probability graph-based transformer state association rule mining method | |
| CN114912077B (en) | Sea wave forecasting method integrating random search and mixed decomposition error correction | |
| CN110263829A (en) | Medium-term and long-term Runoff Forecast method based on CEEMDAN and random forest | |
| CN116861201A (en) | Power supply system based on artificial intelligence | |
| CN114169434A (en) | Load prediction method | |
| CN114492922A (en) | Medium-and-long-term power generation capacity prediction method | |
| CN113409072A (en) | Empirical mode decomposition and distributed GRU neural network and price prediction method | |
| CN110110921A (en) | A kind of river level prediction technique considering time-lag effect | |
| Xu et al. | Short-term wind speed prediction based on GRU | |
| CN113111592A (en) | Short-term wind power prediction method based on EMD-LSTM | |
| CN103209417A (en) | Method and device for predicting spectrum occupancy state based on neural network | |
| CN113128666A (en) | Mo-S-LSTMs model-based time series multi-step prediction method | |
| CN116822743A (en) | Wind power prediction method based on two-stage decomposition reconstruction and error correction | |
| CN115511657A (en) | Wind power output and photovoltaic output evaluation method based on combined prediction model | |
| CN116127833A (en) | Wind power prediction method, system, device and medium based on VMD and LSTM fusion model | |
| CN107967395A (en) | A kind of nonlinear time_varying system Fast Identification Method based on the expansion of beta wavelet basis functions | |
| Zhang et al. | Wind farm wind power prediction method based on CEEMDAN and DE optimized DNN neural network |
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: 20190201 |
|
| RJ01 | Rejection of invention patent application after publication |