CN107203827A - A kind of wind turbine forecasting wind speed optimization method based on multiscale analysis - Google Patents
A kind of wind turbine forecasting wind speed optimization method based on multiscale analysis Download PDFInfo
- Publication number
- CN107203827A CN107203827A CN201710473063.2A CN201710473063A CN107203827A CN 107203827 A CN107203827 A CN 107203827A CN 201710473063 A CN201710473063 A CN 201710473063A CN 107203827 A CN107203827 A CN 107203827A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- mtd
- msup
- sequence
- 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 42
- 238000005457 optimization Methods 0.000 title claims abstract description 39
- 238000004458 analytical method Methods 0.000 title claims abstract description 22
- 239000002245 particle Substances 0.000 claims abstract description 46
- 238000010183 spectrum analysis Methods 0.000 claims abstract description 11
- 238000005516 engineering process Methods 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims description 47
- 238000012360 testing method Methods 0.000 claims description 40
- 230000000737 periodic effect Effects 0.000 claims description 35
- 238000012549 training Methods 0.000 claims description 30
- 230000003595 spectral effect Effects 0.000 claims description 23
- 238000005070 sampling Methods 0.000 claims description 22
- 239000011159 matrix material Substances 0.000 claims description 21
- 238000001228 spectrum Methods 0.000 claims description 15
- 210000002569 neuron Anatomy 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 9
- 230000001133 acceleration Effects 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 238000003062 neural network model Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims 1
- 238000005259 measurement Methods 0.000 abstract description 5
- 238000004422 calculation algorithm Methods 0.000 abstract description 4
- 230000001537 neural effect Effects 0.000 abstract 1
- 238000002474 experimental method Methods 0.000 description 18
- 230000000052 comparative effect Effects 0.000 description 11
- 230000005611 electricity Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 5
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000010248 power generation Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000000053 physical method Methods 0.000 description 3
- 238000007619 statistical method Methods 0.000 description 3
- 238000013277 forecasting method Methods 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011437 continuous method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process 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
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012706 support-vector machine 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
- G06F17/148—Wavelet transforms
-
- 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
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Data Mining & Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Entrepreneurship & Innovation (AREA)
- General Engineering & Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- Algebra (AREA)
- Quality & Reliability (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
Abstract
The invention discloses a kind of wind turbine forecasting wind speed optimization method based on multiscale analysis, its major technique is the principle based on multiscale analysis, using Wavelet spectral analysis technology, extract the harmonic compoment sequence implied in wind turbine measurement wind speed time series and isolated residual sequence, harmonic compoment sequence and residual sequence are predicted respectively using the RBF neural optimized based on particle cluster algorithm, final wind turbine forecasting wind speed result can be obtained by harmonic compoment sequence and predicting the outcome for residual sequence.The present invention is realized carries out fine forecast to the measurement wind speed in wind power plant per typhoon motor, so as to effectively improve the short-term forecast level of exerting oneself of whole wind power plant.
Description
Technical Field
The invention relates to the technical field of wind power generation, in particular to a wind speed prediction optimization method of a wind turbine based on multi-scale analysis.
Background
In order to effectively incorporate wind energy into a power grid, accurate prediction of the output of a wind power plant is extremely necessary and critical, wherein short-term prediction of 0 to 6 hours is of great significance for real-time scheduling of the power grid, ensuring of technical parameters related to power grid safety, such as power grid frequency, power and voltage balance, and the like.
The wind energy is a renewable clean energy, and has the advantages of flexible installation scale, high reliability of the wind power generator set, low manufacturing cost, simple operation and maintenance and the like. According to the monitoring situation of the wind power industry in 2014 published by the national energy agency of 2 months in 2015, the accumulated installed capacity of wind power in China reaches 9637 ten thousand kilowatts, accounts for 7 percent of the installed capacity of all power generation and accounts for 27 percent of the installed capacity of the wind power in the world by the end of 2014. In 2014, the wind power grid electricity quantity is 1534 hundred million kilowatts, which accounts for 2.78% of the total electricity generation quantity. The national energy agency of 12 months in 2014 publishes "strategic action plan for energy development (2014-2020), and the wind power installation is expected to reach 2 hundred million kilowatts in 2020. At present, wind power becomes the third main power supply in China after thermal power and hydroelectric power. With the continuous increase of installed capacity, the problem of electricity abandonment of wind power is always more prominent, and according to the statistics of the national energy agency, the average wind abandonment rate reaches 17% when the electricity abandonment amount of the whole country in 2012 is about 200 hundred million kilowatts; in 2013, the average wind abandon rate reaches 10% when the wind abandon amount of the whole country is about 150 hundred million kilowatts, and the latest statistics show that the average wind abandon rate is 7.5% when the wind abandon amount of the wind power reaches 86 hundred million kilowatts in 2014 to the end of 9 months. One important reason for causing the wind power electricity abandonment is that the wind intermittence causes the fluctuation and instability of the wind power to affect the wind power quality, and the electricity is abandoned in vain for ensuring the safety of a power grid. Based on this, the national energy agency publishes a temporary method for wind power plant power prediction and forecast management in 2011, and requires that all wind power plants which are connected to the grid and operate in China should establish a wind power prediction and forecast system and a power generation plan declaration working mechanism before 1 month and 1 day 2012 and start trial operation, and report a wind power prediction and forecast result according to requirements.
The conventional wind speed forecasting method of the wind power plant comprises a physical method and a statistical method, wherein the physical method is to obtain a timed, fixed-point and quantitative wind power forecasting output result of a numerical weather forecasting mode with high space-time resolution according to a refined numerical weather forecasting mode, and meanwhile, according to the actual operation condition of a wind power plant fan, various fan power generation influence factors are comprehensively considered, an output forecasting physical model is established, and the output forecasting of the wind power plant is carried out. The physical method does not need a large amount of measurement data, but requires accurate mathematical description of the atmospheric physical characteristics and the characteristics of the wind power plant, the equations are difficult to solve, the required data are large in amount, the calculation amount is large, the calculation time is long, and the difficulty and the cost for acquiring the data from a meteorological department are high, so that a statistical method is still commonly used in short-term wind power plant wind speed forecasting. At present, the statistical method mostly adopts a continuous method, a random time sequence method, a Kalman filtering method, a neural network method, a support vector machine and other methods according to historical data of a wind measuring tower of a wind power plant. The biggest disadvantage of forecasting only by relying on anemometer tower data is that a wind power plant is influenced by terrain, turbulence and the like, and the wind speed measured at a wind motor is possibly obviously different from the wind speed measured at the anemometer tower, so that the great forecasting error is inevitably caused by only forecasting the output of the whole wind power plant by using the measured wind speed of the anemometer tower, and the method is irrelevant to a specific forecasting method. With the improvement of the measurement technology and the calculation capability of a computer, the wind speed measured by a single generator can be finely forecasted.
Disclosure of Invention
In view of the above defects in the prior art, the technical problem to be solved by the present invention is to provide a wind turbine wind speed prediction optimization method based on multi-scale analysis, and the wind speed of each wind turbine in a wind farm is finely predicted, so that the short-term output prediction level of the whole wind farm is effectively improved.
The invention relates to a wind speed prediction optimization method of a wind turbine based on multi-scale analysis, which comprises the following steps of:
(1) reading an original sampling wind speed time sequence p of a wind motor, wherein i is 1,2,. Adjusting p to an average wind speed time sequence p '═ { p' (j), j ═ 1,2,. and M } required by a forecast interval, wherein M is the wind turbine generator adjusted according to the forecast interval requirementThe number of sampling points of the average wind speed sequence, the average value of p' isOrder to
(2) Extracting a significant periodic sequence { P implicit in P by adopting a multi-scale wavelet power spectrum analysis technology1,P2,…,Pk,…,PKWhere K is the number of significant periodic sequences implied in P, Pk＝{Pk(1),Pk(2),…,Pk(M), whereby P ═ P }1+P2+…+PK+ R, and R ═ P-P1-P2-…-PKThe residual sequence is the residual sequence after the significant periodic sequence is removed from the P;
(3) for significant periodic sequences in P { P1,P2,…,Pk,…,PKAnd (6) optimizing the RBF neural network by adopting a particle swarm optimization for prediction, and assuming that the prediction step length is l, each significant period sequence { P }1,P2,…,Pk,…,PKThe predicted result is { Y }1,Y2,…,Yk,…,YKIn which Y isk＝{Yk(1),Yk(2),…,Yk(l)}；
(4) The first-order difference sequence D of the residual sequence R is predicted by optimizing the RBF neural network through the particle swarm optimization, and then the prediction result of the residual sequence R is obtained through the difference inverse operation, wherein the assumption is that the prediction step length is l, and the prediction result of the residual sequence R is YR＝{YR(1),YR(2),...,YR(l)}；
(5) Will be provided withAdding the prediction results of the significant periodic sequences and the residual sequence R to obtain a final prediction result Y,
the multi-scale wavelet power spectrum analysis technology comprises the following steps: assume a discrete time series xnAnd N, wherein N is 1, the.. and N, N is total N sampling points, the sampling time interval t is 1, Morlet wavelet transform is applied, the significant periods of the time series are analyzed, and the time series corresponding to each significant period band are extracted.
The multi-scale wavelet power spectrum analysis technology specifically comprises the following steps:
1) determining an analysis period
Period T of wavelet transformjAnd the scale parameter s in wavelet analysisjIn connection, consider the Morlet mother wavelet center period characteristic, here Tj＝sj(ii) a The selection of the scale parameters is: sj＝2jjJ, wherein J is 1/4, J is 0,1,.., J +1 scales, wherein the maximum value of J does not exceed Jmax＝4log2(N) then;
2) determining global wavelet transform spectral values
At the n-th sampling point, sjLocal wavelet transform spectrum value W corresponding to scale parametern(sj) Comprises the following steps:
wherein psi*(. cndot.) is a conjugate function of ψ (. cndot.) and, for the sample point n, the scale parameter sjThe Morlet wavelet basis wavelet function of (a) is:
to Wn(sj) Modulo W ofn(sj) Integrating | along the whole sampling interval to obtain a scale parameter sjCorresponding global wavelet transform spectral valuesNamely:
transforming spectral values using normalized global waveletsWherein sigma2Is xnThe variance of (a);
3) global wavelet transform spectral value significance test
Generally, a period corresponding to a maximum value of a global wavelet transform spectrum value curve is set as a main period, but if the period is significant, a significance test is passed; comparing the obtained global wavelet transform spectrum value with red noise spectrum value to judge its significance, wherein the red noise spectrum value QkExpressed as:
wherein α is xnThe time sequence lags behind the autocorrelation coefficient of one sample point, k being 0, 1.., N/2;
assuming that the global wavelet transform spectral values are some aperiodic process spectral values, the ratio of them to the red noise spectral values follows the one removed by the degree of freedom vDistribution:
in which degree of freedomGamma is a decorrelation factor; therein, theTaking significance level of 0.05 whenThen, the period corresponding to the global wavelet transform spectrum value is significant;
4) extracting time series corresponding to significant periodic bands
Extracting a set period band, i.e., [ T ]1,T2]Corresponding time series x'nThe set period band [ T ] is known from (1)1，T2]Corresponding scale parameter isFor Morlet wavelets, extractionThe time sequence corresponding to the scale parameter corresponds to the scale parameter bandAnd (3) summing, namely:
wherein,is Wn(sj) Real part of for Morlet wavelet, #0(0)＝π-1/4，C ＝0.776。
For the Morlet wavelet, the decorrelation factor γ is 2.32.
The specific process of optimizing the RBF neural network by the particle swarm adopted in the steps (3) and (4) is as follows:
(1-1) determining two parameters to be optimized, wherein the first parameter is the number I of neurons in an input layer of the RBF neural network, and the second parameter is the length L of a training set;
(1-2) initializing group X ═ X1,X2,...,XQ1) Wherein Q is1Is the total number of particles, the ith particle is Xi＝(Ii,Li) Particle velocity of Vi＝(vIi,vLi) In which Ii,LiA set of alternative solutions for parameter I, L;
(1-3) for each particle X in the populationi(Ii,Li) Constructing input and output matrix of RBF neural network training set according to determined parameters, wherein the input and output matrix is used for significant period sequence PkOr residual sequence R and RBF neural network input layer neuron number IiFirst, a matrix Z is established1And Z2Wherein:
here, F denotes a significant periodic sequence PkOr a residual sequence R, for the length L, Z of the neural network training set to be optimized1Middle last LiInput matrix I with columns as training settrain，Z2Middle last LiOutput matrix O with columns as training settrain(ii) a Taking the forecast step length l as the test step length, Z1The last column in the test set is used as an input matrix I of the test settest，Z2The last column in the test set is used as the output matrix O of the test settest(ii) a The method comprises the steps of taking the sum of squares of errors of simulation results of a test set by an RBF neural network constructed according to a training set as a fitness value of the test set, taking the minimum fitness value as an optimization direction as an evaluation standard, evaluating the quality of each particle, and recording the particle XiThe current individual extreme value is Bbest(i) Taking out group Bbest(i) Optimal individual as global extremum Gbest；
(1-4) Each particle X in the populationiUpdating the position and the speed of the mobile terminal respectively;
in the formula: omega is the inertial weight, c1、c2For the acceleration factor, g is the current iteration number, and r1、r2Is distributed in [0,1 ]]The random number of (2);
(1-5) recalculating the objective function value of each particle at that time, and updating Bbest(i) And Gbest；
(1-6) judging whether the maximum iteration times is reached, if so, ending the optimization process, and obtaining a parameter optimal value (I) obtained by particle swarm optimizationbest,Lbest) Otherwise, returning to the step (1-3);
(1-7) according to IbestAnd LbestConstructing RBF neural network training set Z3And test set Z4Wherein:
and establishing an RBF neural network model, performing iterative prediction in step I after training, and obtaining a corresponding prediction result.
The above inertia weight ω0.5, acceleration factor c1＝c2＝1.49445。
The invention has the beneficial effects that:
(1) the obvious periodic sequence of the wind speed measured by the wind turbine and extracted by multi-scale analysis has strong regularity, so that high-precision prediction can be performed, and the obvious periodic sequence has a large proportion in the original wind speed measured by the wind turbine, so that a foundation for high-precision prediction is laid; the residual sequence without the significant periodic sequence has small specific gravity in the wind speed sequence measured by the whole wind turbine on one hand, and becomes stable due to one-time difference operation on the other hand, and the prediction error is relatively limited, so that the original wind speed sequence is decomposed into a plurality of significant periodic sequences and a single residual sequence through multi-scale analysis, and the idea of respectively predicting each significant periodic sequence and each residual sequence can greatly improve the whole prediction effect.
(2) Aiming at the problem that the neural network is easy to fall into overfitting and the problem of determining the number of neurons in an input layer of the RBF neural network, the invention provides a method for establishing particle swarm optimization on the two parameters, so that the generalization performance of the RBF neural network is obviously improved, and the prediction precision is finally improved.
The conception, the specific structure and the technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, the features and the effects of the present invention.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a sequence diagram of average measured wind speed of a wind turbine in the minute level;
FIG. 3 is a graph of the wavelet power spectrum analysis result of P;
FIG. 4 is a graph of a significant periodic sequence of P-extractions and a separated residual sequence;
FIG. 5 is P1Performing a result graph of the 3-step prediction;
FIG. 6 is P2Performing a result graph of the 3-step prediction;
FIG. 7 is P3Performing a result graph of the 3-step prediction;
FIG. 8 is a graph of the 1-step prediction results of the method of the present invention;
FIG. 9 is a graph of 2-step prediction results for the method of the present invention;
FIG. 10 is a graph of the 3-step prediction results of the method of the present invention;
FIG. 11 is a graph of the prediction results of the particle swarm optimization RBF neural network 1 step established in comparative experiment 1;
FIG. 12 is a graph of 2-step prediction results of particle swarm optimization RBF neural network established in comparative experiment 1;
FIG. 13 is a graph of the 3-step prediction results of the particle swarm optimization RBF neural network established in comparative experiment 1;
FIG. 14 is a graph of the 1-step prediction results of the ARIMA time series model established in comparative experiment 2;
FIG. 15 is a 2-step prediction result diagram of the ARIMA time series model established in comparative experiment 2;
FIG. 16 is a 3-step prediction result diagram of the ARIMA time series model established in comparative experiment 2.
Detailed Description
The method of the present invention is illustrated in the flow chart of fig. 1, and includes the following steps.
Firstly, adjusting an original wind speed sequence according to a forecast interval requirement:
reading an original sampling wind speed time sequence p of a wind motor, wherein i is 1,2,. Adjust p to be at forecast intervalsThe required time sequence of the average wind speed p ' ═ { p ' (j), j ═ 1,2,. and M }, wherein M is the number of sampling points of the average wind speed sequence of the wind turbine which is adjusted according to the requirement of the forecast interval, and the average value of p ' isOrder to
Step two, implicit periodic signal extraction:
extracting a significant periodic sequence { P implicit in P by adopting a multi-scale wavelet power spectrum analysis technology1,P2,…,Pk,…,PKWhere K is the number of significant periodic sequences implied in P, Pk＝{Pk(1),Pk(2),…,Pk(M), whereby P ═ P }1+P2+…+PK+ R, and R ═ P-P1-P2-…-PKThe residual sequence is the residual sequence after the significant periodic sequence is removed from the P;
the multi-scale wavelet power spectrum analysis technology comprises the following steps:
assume a discrete time series xnWherein N is 1,.. and N, N is total N sampling points, the sampling time interval t is 1, Morlet wavelet transform is applied, the significant period of the time sequence is analyzed, and the time sequence corresponding to each significant period band is extracted;
1) determining an analysis period
Period T of wavelet transformj(FIG. 3 abscissa value) and the mesoscale parameter s of wavelet analysisjIn connection, consider the Morlet mother wavelet center period characteristic, here Tj＝sj. The selection of the scale parameters is: sj＝2jjJ, wherein J is 1/4, J is 0,1,.., J +1 scales, wherein the maximum value of J does not exceed Jmax＝4log2(N) may be used, where J is 48.
2) Determining global wavelet transform spectral values
At the n-th sampling point, sjLocal wavelet transform spectrum value W corresponding to scale parametern(sj) Comprises the following steps:
wherein psi*(. cndot.) is a conjugate function of ψ (. cndot.) and, for the sample point n, the scale parameter sjThe Morlet wavelet basis wavelet function of (a) is:
to Wn(sj) Modulo W ofn(sj) Integrating | along the whole sampling interval to obtain a scale parameter sjCorresponding global wavelet transform spectral valuesNamely:
the invention uses normalized global wavelet transform spectral values(solid line in FIG. 3), where σ2Is xnThe variance of (c).
3) Global wavelet transform spectral value significance test
The period corresponding to the maximum value of the global wavelet transform spectrum value curve is generally set as a main period, but whether the period is significant or not passes a significance test. The global wavelet transform spectrum value obtained above is compared with red noise spectrum value Q to judge its significancekExpressed as:
wherein α is xnThe time sequence lags behind the autocorrelation coefficient of one sample point, k 0, 1.
Assuming that the global wavelet transform spectral values are some aperiodic process spectral values, the ratio of them to the red noise spectral values follows the one removed by the degree of freedom vDistribution:
in which degree of freedomGamma is the decorrelation factor, and for the Morlet wavelet, gamma is 2.32. Therein, theTaking significance level of 0.05 whenThe period corresponding to the global wavelet transform spectral value is significant, whereinShown as a dashed line in fig. 3.
4) Extracting time series corresponding to significant periodic bands
Extracting a specific periodic band, i.e. [ T ]1,T2]Corresponding time series x'nFrom (1), a specific periodic band [ T ]1，T2]Corresponding scale parameter isFor theMorlet wavelet, extractionThe time sequence corresponding to the scale parameter corresponds to the scale parameter bandAnd (3) summing, namely:
wherein,is Wn(sj) Real part of for Morlet wavelet, #0(0)＝π-1/4，C ＝0.776。
Thirdly, prediction of implicit significant periodic sequences:
for significant periodic sequences in P { P1,P2,…,Pk,…,PKAnd (6) optimizing the RBF neural network by adopting a particle swarm optimization for prediction, and assuming that the prediction step length is l, each significant period sequence { P }1,P2,…,Pk,…,PKThe predicted result is { Y }1,Y2,…,Yk,…,YKIn which Y isk＝{Yk(1),Yk(2),…,Yk(l)}；
Step four, predicting a residual sequence R:
the first-order difference sequence D of the residual sequence R is predicted by optimizing the RBF neural network through the particle swarm optimization, the prediction result of the residual sequence is obtained through the difference inverse operation, and the prediction result of the residual sequence R is Y if the prediction step length is lR＝{YR(1),YR(2),...,YR(l)}；
And step five, obtaining a final prediction result:
will be provided withAdding the prediction results of the significant periodic sequences and the residual sequences to obtain a final prediction result Y,
in this embodiment, the specific process of optimizing the RBF neural network by the particle swarm algorithm adopted in the steps (3) and (4) is as follows:
(1-1) determining 2 parameters to be optimized, wherein the first parameter is the number I of neurons in an input layer of the RBF neural network, and the other parameter is the length L of a training set;
(1-2) initializing populationWherein Q1Is the total number of particles, the ith particle is Xi＝(Ii,Li) The particle velocity isWherein Ii,LiA set of alternative solutions for parameter I, L;
(1-3) for each particle X in the populationi(Ii,Li) Constructing input and output matrix of RBF neural network training set according to determined parameters, wherein the input and output matrix is used for significant period sequence PkOr residual sequence R and RBF neural network input layer neuron number IiFirst, a matrix Z is established1And Z2Wherein:
here, F denotes a significant periodic sequence PkOr a residual sequence R, for the length L, Z of the neural network training set to be optimized1Middle last LiInput matrix I with columns as training settrain，Z2Middle last LiOutput matrix O with columns as training settrain(ii) a Taking the forecast step length l as the test step length, Z1The last column in the test set is used as an input matrix I of the test settest，Z2The last column in the test set is used as the output matrix O of the test settest(ii) a The method comprises the steps of taking the sum of squares of errors of simulation results of a test set by an RBF neural network constructed according to a training set as a fitness value of the test set, taking the minimum fitness value as an optimization direction as an evaluation standard, evaluating the quality of each particle, and recording the particle XiThe current individual extreme value is Bbest(i) Taking out group Bbest(i) Optimal individual as global extremum Gbest；
(1-4) Each particle X in the populationiUpdating the position and the speed of the mobile terminal respectively;
in the formula: omega is the inertial weight, c1、c2For the acceleration factor, usually ω is 0.5, c1＝c21.49445, g is the current iteration number, and r1、r2Is distributed in [0,1 ]]The random number of (2);
(1-5) recalculating the objective function value of each particle at that time, and updating Bbest(i) And Gbest；
(1-6) judging whether the maximum iteration times is reached, if so, ending the optimization process, and obtaining a parameter optimal value (I) obtained by particle swarm optimizationbest,Lbest) Otherwise, returning to the step (1-3);
(1-7) according to IbestAnd LbestConstructing RBF neural network training set Z3And test set Z4Wherein:
and establishing an RBF neural network model, performing iterative prediction in step I after training, and obtaining a corresponding prediction result.
Specific test examples:
according to the flow chart shown in figure 1, a wind motor measurement wind speed time sequence collected from a wind motor 1# in a certain wind power plant in China from 10 months, 5 days and 10 days in 2015 to 07 minutes and 53 seconds is taken, and due to the ultra-short-term forecast of the minute level shown in the example, firstly according to the forecast interval requirement, according to the forecast interval requirementAdjusting the wind speed time sequence measured by the wind turbine to obtain an average wind speed time sequence in the level of minutes, as shown in fig. 2, where p (i) is the measured wind speed time sequence (in the level of seconds, but the sampling intervals are not uniform) collected by the wind turbine, p '(i) is the average wind speed time sequence in the level of minutes after adjustment, and t' respectively refer to the serial numbers of the sampling points corresponding to the start and stop points of the minutes in the original wind speed time sequence. The test example takes the first 2310 data in p' (i) as a known data set, and carries out 1-step, 2-step and 3-step prediction experiments for the next 100 steps to examine the effectiveness of the algorithm.
The effectiveness of the algorithm is tested by taking the mean square error MSE, the relative percentage error MAPE and the average absolute error MAE as standards respectively:
wherein Y (i) and p' (i) are respectively a predicted value and a true value of the wind speed measured by the wind turbine, and l is a prediction step length.
The wind turbine average wind speed sequence of the minute level after the average value is removed is marked as P, and fig. 3 shows the wavelet power spectrum analysis result of P, and the wind turbine wind speed sequence is found to have a significant period with 3 sampling points of 4096, 1217 and 609 as extreme points, and the first period points lower than the red noise detection line on the left and right sides of the extreme points are taken to form a period band, which is a significant period band, in this example, 3 significant period bands are [2896,4871 ]]、[861,1722]And [512,724]Extracting time sequences corresponding to the 3 periodic bands respectively as P according to a wavelet reconstruction method1、P2And P3And obtaining a corresponding residual sequence R, whereby P ═ P1+P2+P3+ R, see FIG. 4. Therefore, the 3 significant periodic sequences have extremely strong regularity and can be predicted with higher precision; on the other hand, although the prediction error for the residual is unavoidable, it is calculated that the energy (variance) of the residual R accounts for 46.61% of the energy (variance) of P, and therefore the prediction error for the residual is much smaller than the error of directly predicting P.
Although the neural network has strong nonlinear fitting capability and quick learning capability, how to construct a training set and a testing set of the neural network still mainly depends on manual experience or trial and error, and the universality is poor.
To P1、P2And P3Adopting a particle swarm optimization-based RBF neural network model, and taking the range of the number of neurons in an input layer as [2,30]]The length of the training set is [100,2200]]The population size of the particle swarm is 50, iteration is performed for 30 times, and Table 1 shows that when 3-step prediction is performed, a significant periodic signal P is aimed at1、P2、P3And the optimization results of two parameters of the number I of input layer neurons of the residual error R and the length L of the training set are as follows:
TABLE 1 pairs of P1、P2、P3Neural network parameter optimization results of sum R
| P1 | P2 | P3 | R | |
| Input layer neuron number (I) | 17 | 15 | 27 | 11 |
| Training set length (L) | 324 | 120 | 177 | 326 |
The test case carries out 1-step, 2-step and 3-step prediction experiments with a total prediction step length of 100, and figures 5, 6 and 7 are specific to P1、P2And P3As a result of performing the 3-step prediction, the error is shown in table 2, and it can be seen that the prediction error for the significant periodic sequence is small, and the overall error is mainly generated due to the residual sequence.
TABLE 2P1、P2And P3Error of performing 3-step prediction
| P1 | P2 | P3 | |
| MSE | 8.4059e-08 | 2.1534e-07 | 2.4510e-07 |
| MAPE | 6.7380e-08 | 1.6800e-07 | 1.5535e-07 |
| MAE | 6.6091e-08 | 1.6513e-07 | 1.7066e-07 |
The total prediction results are shown in fig. 8, 9 and 10, and table 3 shows the statistics of prediction errors, and it can be seen that, except for the first 10 prediction step errors, the subsequent prediction result errors are all smaller, and the total prediction results are more satisfactory.
TABLE 3 prediction error of the patented method
| 1 step prediction | 2-step prediction | 3-step prediction | |
| MSE | 0.4572 | 0.6077 | 0.8274 |
| MAPE | 0.0270 | 0.0348 | 0.0461 |
| MAE | 0.3934 | 0.5054 | 0.6715 |
Comparative experiment 1
In order to verify the influence of the differential processing means provided by the patent and the optimization of two parameters, namely the neuron number I of the input layer of the neural network and the length L of the training set, on the experimental result, a differential operation is directly performed on the wind speed sequence p' of the wind turbine in comparison with experiment 1, then the RBF neural network optimized by the particle swarm optimization is established, the range of the neuron number of the input layer is taken as [2,30], the length of the training set is taken as [100,2200], the particle swarm size is taken as 50, and the iteration is performed for 30 times. Table 4 shows the results of the RBF neural network parameter optimization established for p' when performing the 3-step prediction.
TABLE 4 RBF neural network parameter optimization results for original sequence p
Similarly, comparative experiment 1 performed 1-step, 2-step and 3-step prediction experiments with a total prediction step size of 100, the prediction results are shown in fig. 11, 12 and 13, table 4 is the statistics of prediction errors, and as can be seen from comparative table 2, the 3 error indicators of the prediction errors are increased by 141.33%, 96.46% and 98.74% respectively with respect to table 2.
TABLE 5 particle swarm optimization RBF neural network prediction error established for original sequence p
| 1 step prediction | 2-step prediction | 3-step prediction | |
| MSE | 1.1253 | 1.6916 | 2.5834 |
| MAPE | 0.0541 | 0.0773 | 0.1099 |
| MAE | 0.7968 | 1.1355 | 1.6272 |
If the difference operation is not performed on p', the number I of neurons in the input layer of the RBF neural network and the length L of the training set are randomly selected, the final prediction error difference is large, and the influence of two groups of different I and L on the final prediction error is selected and explained, as shown in Table 5.
TABLE 6 RBF neural network prediction error for original sequence p
The 3 error indexes of the comparative experiments of two groups of different parameters are respectively increased by 169.67%, 155.31%, 162.75% and 138.51%, 126.97% and 133.15% in the table 2. The poor effect of the group on the comparison experiment shows that the selection of the neural network parameters has great influence on the learning ability and generalization of the neural network, so that the effect of directly adopting the neural network for modeling is not good.
Comparative experiment 2
A differential autoregressive moving average model (ARIMA) was built for p'. Selecting 100 sampling data points before the prediction point, determining the structure of the ARIMA model by an AIC criterion order-fixing method, similarly, carrying out 1-step, 2-step and 3-step prediction experiments with a total prediction step length of 100 in a comparison experiment 2, wherein the prediction results are shown in figures 14, 15 and 16, a table 7 is prediction error statistics, and 3 error indexes are respectively increased by 18.93%, 9.54% and 8.77% in the comparison table 2.
TABLE 7 ARIMA time series model prediction error for p' build
| 1 step prediction | 2-step prediction | 3-step prediction | |
| MSE | 0.6761 | 0.7934 | 0.8926 |
| MAPE | 0.0346 | 0.0407 | 0.0458 |
| MAE | 0.5015 | 0.5883 | 0.6572 |
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (6)
1. A wind turbine wind speed prediction optimization method based on multi-scale analysis is characterized by comprising the following steps:
(1) reading an original sampling wind speed time sequence p of a wind motor, wherein i is 1,2,. Adjusting p to an average wind speed time sequence p ' ═ { p ' (j), j ═ 1,2,. and M } required by a forecast interval, wherein M is the number of sampling points of the average wind speed sequence of the wind turbine adjusted according to the forecast interval, and the average value of p ' isOrder to
(2) Extracting a significant periodic sequence { P implicit in P by adopting a multi-scale wavelet power spectrum analysis technology1,P2,…,Pk,…,PKWhere K is the number of significant periodic sequences implied in P, Pk＝{Pk(1),Pk(2),…,Pk(M), whereby P ═ P }1+P2+…+PK+ R, and R ═ P-P1-P2-…-PKThe residual sequence is the residual sequence after the significant periodic sequence is removed from the P;
(3) for significant periodic sequences in P { P1,P2,…,Pk,…,PKAnd (6) optimizing the RBF neural network by adopting a particle swarm optimization for prediction, and assuming that the prediction step length is l, each significant period sequence { P }1,P2,…,Pk,…,PKThe predicted result is { Y }1,Y2,…,Yk,…,YKIn which Y isk＝{Yk(1),Yk(2),…,Yk(l)}；
(4) The first-order difference sequence D of the residual sequence R is predicted by optimizing the RBF neural network through the particle swarm optimization, and then the prediction result of the residual sequence R is obtained through the difference inverse operation, wherein the assumption is that the prediction step length is l, and the prediction result of the residual sequence R is YR＝{YR(1),YR(2),...,YR(l)}；
(5) Will be provided withAdding the prediction results of the significant periodic sequences and the residual sequence R to obtain a final prediction result Y,
2. the method of claim 1The wind speed prediction optimization method based on multi-scale analysis is characterized in that the multi-scale wavelet power spectrum analysis technology comprises the following steps: assume a discrete time series xnAnd N, wherein N is 1, the.. and N, N is total N sampling points, the sampling time interval t is 1, Morlet wavelet transform is applied, the significant periods of the time series are analyzed, and the time series corresponding to each significant period band are extracted.
3. The wind turbine wind speed prediction optimization method based on multi-scale analysis according to claim 2, wherein the multi-scale wavelet power spectrum analysis technique specifically comprises the following steps:
1) determining an analysis period
Period T of wavelet transformjAnd the scale parameter s in wavelet analysisjIn connection, consider the Morlet mother wavelet center period characteristic, here Tj＝sj(ii) a The selection of the scale parameters is: sj＝2jjJ, wherein J is 1/4, J is 0,1,.., J +1 scales, wherein the maximum value of J does not exceed Jmax＝4log2(N) then;
2) determining global wavelet transform spectral values
At the n-th sampling point, sjLocal wavelet transform spectrum value W corresponding to scale parametern(sj) Comprises the following steps:
<mrow> <msub> <mi>W</mi> <mi>n</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>s</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mrow> <mo>(</mo> <mfrac> <mn>1</mn> <msub> <mi>s</mi> <mi>j</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mrow> <mn>1</mn> <mo>/</mo> <mn>2</mn> </mrow> </msup> <munderover> <mo>&Sigma;</mo> <mrow> <msup> <mi>n</mi> <mo>&prime;</mo> </msup> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msub> <mi>x</mi> <msup> <mi>n</mi> <mo>&prime;</mo> </msup> </msub> <msup> <mi>&psi;</mi> <mo>*</mo> </msup> <mrow> <mo>(</mo> <mfrac> <mrow> <msup> <mi>n</mi> <mo>&prime;</mo> </msup> <mo>-</mo> <mi>n</mi> </mrow> <msub> <mi>s</mi> <mi>j</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>
wherein psi*(. cndot.) is a conjugate function of ψ (. cndot.) and, for the sample point n, the scale parameter sjThe Morlet wavelet basis wavelet function of (a) is:
<mrow> <mi>&psi;</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <msup> <mi>n</mi> <mo>&prime;</mo> </msup> <mo>-</mo> <mi>n</mi> </mrow> <msub> <mi>s</mi> <mi>j</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>&pi;</mi> <mrow> <mo>-</mo> <mn>1</mn> <mo>/</mo> <mn>4</mn> </mrow> </msup> <msup> <mi>e</mi> <mrow> <mi>i</mi> <mn>2</mn> <mi>&pi;</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <msup> <mi>n</mi> <mo>&prime;</mo> </msup> <mo>-</mo> <mi>n</mi> </mrow> <msub> <mi>s</mi> <mi>j</mi> </msub> </mfrac> <mo>)</mo> </mrow> </mrow> </msup> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mrow> <mo>(</mo> <mfrac> <mrow> <msup> <mi>n</mi> <mo>&prime;</mo> </msup> <mo>-</mo> <mi>n</mi> </mrow> <msub> <mi>s</mi> <mi>j</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>
to Wn(sj) Modulo W ofn(sj) Integrating | along the whole sampling interval to obtain a scale parameter sjCorresponding global wavelet transform spectral valuesNamely:
<mrow> <msup> <mover> <mi>W</mi> <mo>&OverBar;</mo> </mover> <mn>2</mn> </msup> <mrow> <mo>(</mo> <msub> <mi>s</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mi>N</mi> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>n</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mo>|</mo> <msub> <mi>W</mi> <mi>n</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>s</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <msup> <mo>|</mo> <mn>2</mn> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow>
transforming spectral values using normalized global waveletsWherein sigma2Is xnThe variance of (a);
3) global wavelet transform spectral value significance test
Transforming the global wavelet into the maxima of the spectral value curveThe period corresponding to the value is determined as a main period, but if the period is significant, the significance test is passed; comparing the obtained global wavelet transform spectrum value with red noise spectrum value to judge its significance, wherein the red noise spectrum value QkExpressed as:
<mrow> <msub> <mi>Q</mi> <mi>k</mi> </msub> <mo>=</mo> <mfrac> <mrow> <mn>1</mn> <mo>-</mo> <msup> <mi>&alpha;</mi> <mn>2</mn> </msup> </mrow> <mrow> <mn>1</mn> <mo>+</mo> <msup> <mi>&alpha;</mi> <mn>2</mn> </msup> <mo>-</mo> <mn>2</mn> <mi>&alpha;</mi> <mi>c</mi> <mi>o</mi> <mi>s</mi> <mrow> <mo>(</mo> <mn>2</mn> <mi>&pi;</mi> <mi>k</mi> <mo>/</mo> <mi>N</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>
wherein α is xnThe time sequence lags behind the autocorrelation coefficient of one sample point, k being 0, 1.., N/2;
assuming that the global wavelet transform spectral values are some aperiodic process spectral values, the ratio of them to the red noise spectral values follows the one removed by the degree of freedom vDistribution:
<mrow> <mfrac> <mrow> <msup> <mover> <mi>W</mi> <mo>&OverBar;</mo> </mover> <mn>2</mn> </msup> <mrow> <mo>(</mo> <msub> <mi>s</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <msup> <mi>&sigma;</mi> <mn>2</mn> </msup> <msub> <mi>Q</mi> <mi>k</mi> </msub> </mrow> </mfrac> <mo>=</mo> <mfrac> <msubsup> <mi>&chi;</mi> <mi>&nu;</mi> <mn>2</mn> </msubsup> <mi>&nu;</mi> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>
in which degree of freedomGamma is a decorrelation factor; therein, theTaking significance level of 0.05 whenThen, the period corresponding to the global wavelet transform spectrum value is significant;
4) extracting time series corresponding to significant periodic bands
Extracting a set period band, i.e., [ T ]1,T2]Corresponding time series x'nThe set period band [ T ] is known from (1)1，T2]Corresponding scale parameter isFor Morlet wavelets, extractionThe time sequence corresponding to the scale parameter corresponds to the scale parameter bandAnd (3) summing, namely:
wherein,is Wn(sj) Real part of for Morlet wavelet, #0(0)＝π-1/4，C ＝0.776。
4. The wind turbine wind speed prediction optimization method based on multi-scale analysis according to claim 3, characterized in that: for the Morlet wavelet, the decorrelation factor γ is 2.32.
5. The wind turbine wind speed prediction optimization method based on multi-scale analysis according to claim 1, characterized in that: the specific process of optimizing the RBF neural network by the particle swarm adopted in the steps (3) and (4) is as follows:
(1-1) determining two parameters to be optimized, wherein the first parameter is the number I of neurons in an input layer of the RBF neural network, and the second parameter is the length L of a training set;
(1-2) initializing populationWherein Q1Is the total number of particles, the ith particle is Xi＝(Ii,Li) The particle velocity isWherein Ii,LiA set of alternative solutions for parameter I, L;
(1-3) for each particle X in the populationi(Ii,Li) Constructing input and output matrix of RBF neural network training set according to determined parameters, wherein the input and output matrix is used for significant period sequence PkOr residual sequence R and RBF neural network input layer neuron number IiFirst, a matrix Z is established1And Z2Wherein:
<mrow> <msub> <mi>Z</mi> <mn>1</mn> </msub> <mo>=</mo> <msub> <mfenced open = "[" close = "]"> <mtable> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>...</mn> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mn>...</mn> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> <mrow> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>*</mo> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </msub> <mo>,</mo> </mrow>
<mrow> <msub> <mi>Z</mi> <mn>2</mn> </msub> <mo>=</mo> <msub> <mfenced open = "[" close = "]"> <mtable> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> <mo>+</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> <mrow> <mi>M</mi> <mo>-</mo> <msub> <mi>I</mi> <mi>i</mi> </msub> </mrow> </msub> </mrow>
here, F denotes a significant periodic sequence PkOr a residual sequence R, for the length L, Z of the neural network training set to be optimized1Middle last LiInput matrix I with columns as training settrain，Z2Middle last LiOutput matrix O with columns as training settrain(ii) a Taking the forecast step length l as the test step length, Z1The last column in the test set is used as an input matrix I of the test settest，Z2The last column in the test set is used as the output matrix O of the test settest(ii) a The method comprises the steps of taking the sum of squares of errors of simulation results of a test set by an RBF neural network constructed according to a training set as a fitness value of the test set, taking the minimum fitness value as an optimization direction as an evaluation standard, evaluating the quality of each particle, and recording the particle XiThe current individual extreme value is Bbest(i) Taking out group Bbest(i) Optimal individual as global extremum Gbest；
(1-4) Each particle X in the populationiUpdating the position and the speed of the mobile terminal respectively;
<mrow> <msubsup> <mi>V</mi> <mi>i</mi> <mrow> <mi>g</mi> <mo>+</mo> <mn>1</mn> </mrow> </msubsup> <mo>=</mo> <msubsup> <mi>&omega;V</mi> <mi>i</mi> <mi>g</mi> </msubsup> <mo>+</mo> <msub> <mi>c</mi> <mn>1</mn> </msub> <msub> <mi>r</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <msub> <mi>B</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>(</mo> <mi>i</mi> <mo>)</mo> <mo>-</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mi>g</mi> </msubsup> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>c</mi> <mn>2</mn> </msub> <msub> <mi>r</mi> <mn>2</mn> </msub> <mrow> <mo>(</mo> <msub> <mi>G</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>-</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mi>g</mi> </msubsup> <mo>)</mo> </mrow> <mo>,</mo> </mrow>
<mrow> <msubsup> <mi>X</mi> <mi>i</mi> <mrow> <mi>g</mi> <mo>+</mo> <mn>1</mn> </mrow> </msubsup> <mo>=</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mi>g</mi> </msubsup> <mo>+</mo> <msubsup> <mi>V</mi> <mi>i</mi> <mrow> <mi>g</mi> <mo>+</mo> <mn>1</mn> </mrow> </msubsup> </mrow>
in the formula: omega is the inertial weight, c1、c2For the acceleration factor, g is the current iteration number, and r1、r2Is distributed in [0,1 ]]The random number of (2);
(1-5) recalculating the objective function value of each particle at that time, and updating Bbest(i) And Gbest；
(1-6) judging whether the maximum iteration times is reached, if so, ending the optimization process, and obtaining a parameter optimal value (I) obtained by particle swarm optimizationbest,Lbest) Otherwise, returning to the step (1-3);
(1-7) according to IbestAnd LbestConstruction of RBF neural network trainingCollection Z3And test set Z4Wherein:
<mrow> <msub> <mi>Z</mi> <mn>3</mn> </msub> <mo>=</mo> <msub> <mfenced open = "[" close = "]"> <mtable> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>...</mn> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mn>...</mn> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> <mrow> <msub> <mi>I</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>*</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> </mrow> </msub> <mo>,</mo> </mrow>
<mrow> <msub> <mi>Z</mi> <mn>4</mn> </msub> <mo>=</mo> <msub> <mfenced open = "[" close = "]"> <mtable> <mtr> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>-</mo> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>+</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>M</mi> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> <msub> <mi>L</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> </msub> </mrow>
and establishing an RBF neural network model, performing iterative prediction in step I after training, and obtaining a corresponding prediction result.
6. The wind turbine wind speed prediction optimization method based on multi-scale analysis according to claim 5, characterized in that: the inertia weight ω is 0.5, and the acceleration factor c1＝c2＝1.49445。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710473063.2A CN107203827A (en) | 2017-06-21 | 2017-06-21 | A kind of wind turbine forecasting wind speed optimization method based on multiscale analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710473063.2A CN107203827A (en) | 2017-06-21 | 2017-06-21 | A kind of wind turbine forecasting wind speed optimization method based on multiscale analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107203827A true CN107203827A (en) | 2017-09-26 |
Family
ID=59907980
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710473063.2A Pending CN107203827A (en) | 2017-06-21 | 2017-06-21 | A kind of wind turbine forecasting wind speed optimization method based on multiscale analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107203827A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109002860A (en) * | 2018-07-27 | 2018-12-14 | 中南大学 | A kind of line of high-speed railway mutation wind speed intelligence adaptability matching prediction technique |
| CN110566405A (en) * | 2019-08-29 | 2019-12-13 | 北京金风科创风电设备有限公司 | power optimization method and device of wind generating set |
| CN111562599A (en) * | 2020-05-14 | 2020-08-21 | 中国北方工业有限公司 | Method for establishing RTD pseudo-range correction forecasting model of Beidou satellite network |
| CN114492090A (en) * | 2022-04-12 | 2022-05-13 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Road surface temperature short-term forecasting method |
| CN114493004A (en) * | 2022-01-27 | 2022-05-13 | 国网山东省电力公司济南供电公司 | Machine learning-based short-term and temporary wind speed extrapolation method for single-point wind speed |
-
2017
- 2017-06-21 CN CN201710473063.2A patent/CN107203827A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109002860A (en) * | 2018-07-27 | 2018-12-14 | 中南大学 | A kind of line of high-speed railway mutation wind speed intelligence adaptability matching prediction technique |
| CN109002860B (en) * | 2018-07-27 | 2020-11-24 | 中南大学 | Intelligent adaptive matching prediction method for sudden change wind speed along high-speed railway |
| CN110566405A (en) * | 2019-08-29 | 2019-12-13 | 北京金风科创风电设备有限公司 | power optimization method and device of wind generating set |
| CN111562599A (en) * | 2020-05-14 | 2020-08-21 | 中国北方工业有限公司 | Method for establishing RTD pseudo-range correction forecasting model of Beidou satellite network |
| CN114493004A (en) * | 2022-01-27 | 2022-05-13 | 国网山东省电力公司济南供电公司 | Machine learning-based short-term and temporary wind speed extrapolation method for single-point wind speed |
| CN114493004B (en) * | 2022-01-27 | 2024-01-12 | 国网山东省电力公司济南供电公司 | Single-point wind speed short-term wind speed extrapolation method based on machine learning |
| CN114492090A (en) * | 2022-04-12 | 2022-05-13 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Road surface temperature short-term forecasting method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107203827A (en) | A kind of wind turbine forecasting wind speed optimization method based on multiscale analysis | |
| Qu et al. | Multi-step wind speed forecasting based on a hybrid decomposition technique and an improved back-propagation neural network | |
| CN102663513B (en) | Utilize the wind power combined prediction modeling method of grey relational grade analysis | |
| Yang et al. | Some issues in uncertainty quantification and parameter tuning: A case study of convective parameterization scheme in the WRF regional climate model | |
| CN102184337B (en) | Dynamic combination analysis method of new energy generating capacity influenced by meteorological information | |
| CN103279804B (en) | The Forecasting Methodology of super short-period wind power | |
| CN109492823B (en) | Method for predicting icing thickness of power transmission line | |
| CN111160520A (en) | BP neural network wind speed prediction method based on genetic algorithm optimization | |
| CN112686464A (en) | Short-term wind power prediction method and device | |
| CN103996084B (en) | Wind power probability Forecasting Methodology based on longitudinal moment Markov chain model | |
| CN107704953A (en) | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests | |
| CN108388962B (en) | Wind power prediction system and method | |
| CN107330183A (en) | A kind of wind power utilization computational methods based on service data | |
| CN107301475A (en) | Load forecast optimization method based on continuous power analysis of spectrum | |
| CN103559561A (en) | Super-short-term prediction method of photovoltaic power station irradiance | |
| CN108717579B (en) | Short-term wind power interval prediction method | |
| CN106229972A (en) | A kind of integrated based on many meteorological sources and the wind power forecasting method of segmentation modeling | |
| CN107169612A (en) | The prediction of wind turbine active power and error revising method based on neutral net | |
| CN103473621A (en) | Wind power station short-term power prediction method | |
| CN110991725B (en) | RBF ultra-short-term wind power prediction method based on wind speed frequency division and weight matching | |
| CN106203693A (en) | A kind of system and method for Power Output for Wind Power Field climbing event prediction | |
| CN105356462A (en) | Wind farm harmonic evaluation prediction method | |
| CN111091236A (en) | Multi-classification deep learning short-term wind power prediction method classified according to pitch angles | |
| Lodge et al. | Short term wind speed prediction using artificial neural networks | |
| CN112633565A (en) | Photovoltaic power aggregation interval prediction method |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170926 |