CN114202107B - Ultra-short-term wind speed prediction method and device - Google Patents
Ultra-short-term wind speed prediction method and device Download PDFInfo
- Publication number
- CN114202107B CN114202107B CN202111370814.0A CN202111370814A CN114202107B CN 114202107 B CN114202107 B CN 114202107B CN 202111370814 A CN202111370814 A CN 202111370814A CN 114202107 B CN114202107 B CN 114202107B
- Authority
- CN
- China
- Prior art keywords
- tensor
- order
- sequence
- wind speed
- time
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 96
- 230000009466 transformation Effects 0.000 claims abstract description 41
- 238000005070 sampling Methods 0.000 claims abstract description 22
- 230000015654 memory Effects 0.000 claims abstract description 15
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 34
- 238000000354 decomposition reaction Methods 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000010076 replication Effects 0.000 claims description 11
- 230000009467 reduction Effects 0.000 claims description 5
- 238000010183 spectrum analysis Methods 0.000 claims description 5
- 230000007787 long-term memory Effects 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 230000003190 augmentative effect Effects 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 3
- 230000008521 reorganization Effects 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 6
- 238000007619 statistical method Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 238000006467 substitution reaction Methods 0.000 description 3
- 229910052623 talc Inorganic materials 0.000 description 3
- 235000012222 talc Nutrition 0.000 description 3
- 239000000454 talc Substances 0.000 description 3
- 241001556567 Acanthamoeba polyphaga mimivirus Species 0.000 description 2
- 241001123248 Arma Species 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000009194 climbing Effects 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 2
- 230000008707 rearrangement Effects 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 241000228740 Procrustes Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2462—Approximate or statistical queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2474—Sequence data queries, e.g. querying versioned data
-
- 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)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Probability & Statistics with Applications (AREA)
- Tourism & Hospitality (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Fuzzy Systems (AREA)
- General Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- General Health & Medical Sciences (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a device for predicting ultra-short-term wind speed, wherein the method comprises the following steps: acquiring wind speed time data, and sampling the wind speed time data at intervals to acquire a time sequence; decomposing the time sequence obtained by sampling into an approximate sequence and a detail sequence; performing Hakking treatment on the high-order tensors of the approximate sequence and the detail sequence to obtain a third-order tensor; slicing the third-order tensor along the time dimension, and decomposing the second-order tensor slice at each moment by using a Tack to obtain a low-rank high-order time sequence; the core tensor time sequence is sent to a long-period memory network for learning, optimal network parameters are obtained through training, and second-order core tensor slices at future time are obtained through prediction; and carrying out transformation processing on the second-order core tensor slice to obtain a future wind speed predicted value. The method comprises the steps of firstly decomposing a one-dimensional wind speed time sequence in a data processing link to obtain high-frequency characteristics and low-frequency characteristics; the prediction capability is effectively improved, and the method can be widely applied to the field of wind speed prediction of wind power plants.
Description
Technical Field
The invention relates to the field of wind speed prediction and economic dispatch of wind power plants, in particular to an ultra-short-term wind speed prediction method and device.
Background
With the development of wind power technology, wind energy has become one of the most potential renewable energy sources. However, due to the nonlinearity, intermittence, and chaos of wind energy, large scale grid connection of wind energy can have a great impact on the safety and stability of the power system, creating difficulties for energy management and economic dispatch. If the predicted value of the wind energy is higher than the actual value, the rotation reserve capacity in the power system is insufficient, so that the system cannot be regulated to a steady-state operation mode; if the predicted value of wind energy is lower than the actual value, the backup capacity of the system may be too sufficient, resulting in an increase in the economic cost of the system. In order to effectively solve the problem, students have conducted a great deal of research in the field of wind speed prediction, and high-precision wind speed prediction can provide effective reference for wind power plant energy scheduling management. Therefore, the model capable of accurately predicting the wind speed change trend is of great significance.
There are two parallel methods in the history of wind speed prediction, namely physical and statistical (including machine learning) methods. Compared with a physical method, the statistical method is more suitable for ultra-short-term wind speed prediction, but the existing prediction model has the following problems:
(1) Machine learning class statistical methods represented by Extreme Learning Machines (ELMs), automatic encoders, and recurrent neural networks. These methods are popular because they have better accuracy or faster computation speed than other methods. However, for wind speed data with strong nonlinearity, the prediction accuracy is still limited because the fitting capability of the numerical model itself is always limited.
(2) In recent years, hybrid models have been increasingly considered because they have been found to maximize the advantages of the above methods, expanding the model fitting capability. However, this kind of model tends to ignore the characteristics of the wind speed data itself, and the interpretation of this kind of model is relatively weak.
(3) In addition, some new works are also proposed to realize high-dimensional time series prediction, try to mine the non-linear characteristic of the data from high dimensions, improve the fitting capacity of the model and strengthen the interpretability of the model. Taking into account the nonlinear characteristics of the compressed wind speed, a high-dimensional embedding method can be used. These methods can improve the prediction performance, but the computational complexity is high.
Disclosure of Invention
In order to solve at least one of the technical problems existing in the prior art to a certain extent, the invention aims to provide an ultra-short-term wind speed prediction method and device.
The technical scheme adopted by the invention is as follows:
an ultra-short term wind speed prediction method comprising the steps of:
acquiring wind speed time data, and sampling the wind speed time data at intervals to acquire a time sequence;
decomposing the time sequence obtained by sampling into an approximate sequence and a detail sequence;
Performing Hakking processing on the high-order tensors of the approximate sequence and the detail sequence to obtain a third-order tensor corresponding to the approximate sequence Third-order tensor/>, corresponding to detail sequence
For third-order tensorAnd third order tensor/>Slicing along the time dimension, and decomposing the second-order tensor slice at each moment by using a Tak to obtain a low-rank high-order time sequence/>, which corresponds to the approximate sequenceLow-rank high-order time sequence/>, corresponding to detail sequence
Sampling two long-short-time memory networks to respectively carry out low-rank high-order time sequenceAnd low rank high order time seriesLearning and prediction are carried out, and core tensor/> is obtained through predictionAnd core tensor/>
The core tensorAnd core tensor/>Reduction is carried out and spliced to a time sequence/>And time series/>In (c) obtaining tensor/>Sum tensor/>For tensor/>Sum tensor/>Performing inverse hank transformation to obtain a predicted approximate wind speed f (N 0 +1) and a detail wind speed s (N 0 +1), and obtaining a predicted wind speed according to the approximate wind speed f (N 0 +1) and the detail wind speed s (N 0 +1).
Further, the Hakking process is performed on the higher-order tensors of the approximate sequence and the detail sequence to obtain a third-order tensor corresponding to the approximate sequenceThird-order tensor/>, corresponding to detail sequenceComprising the following steps:
Regarding the approximate sequence F (t) and the detail sequence S (t) as a first-order tensor F and a first-order tensor S with the size of 1 XN 0, and performing one-way delay embedding transformation to obtain a second-order tensor F and a second-order tensor S;
dividing a second-order tensor F along a second dimension to obtain a plurality of first-order tensor slices F (t) at different moments, calculating Euclidean distances between the plurality of first-order tensor slices F (t) and a sample to be predicted, dividing M first-order tensor slices with the minimum Euclidean distance into the same sample space, and rearranging the first-order tensor slices in the sample space into second-order tensors FR according to the sequence from the large distance to the small distance;
The second-order tensor FR and the second-order tensor S are subjected to one-way delay embedded transformation to obtain a third-order tensor corresponding to the approximate sequence Third-order tensor/>, corresponding to detail sequence
Further, the treating the approximation sequence F (t) as a first-order tensor F with the size of 1×n 0, performing a one-way delay embedding transformation to obtain a second-order tensor F, including:
Considering the approximation sequence f (t) as a first order tensor f, f= [ f (1), f (2) ], for a1×n 0 size, f (t), f (N 0), time t=1, 2, …, N 0;
Selecting the delay time tau 1 of the embedding transformation to construct a unique replica matrix
Performing single-way delay embedded transformation to obtain a second-order tensor F, wherein the single-way delay embedded transformation is expressed as the result of linear tensor sum-folding operation of the first-order tensor F and the Mode-1 of the unique replication matrix D 1;
The form of the approximate second order tensor F obtained after folding is as follows:
Further, the dividing the second-order tensor F along the second dimension to obtain a plurality of first-order tensor slices F (t) at different moments, calculating the euclidean distance between the plurality of first-order tensor slices F (t) and the sample to be predicted, dividing M first-order tensor slices with the minimum euclidean distance into the same sample space, and rearranging the first-order tensor slices in the sample space into second-order tensors FR according to the sequence from the larger distance to the smaller distance, including:
Slicing the approximate second-order tensor F along the second dimension at time to obtain (N 0-τ1 +1) first-order tensor slices F (t) = [ F (t) F (t+1) … F (t+τ 1)]T,t=1,2,...,N0-τ1 +1;
based on Euclidean distance theory, F (t) at different moments and F (N 0-τ1 +1) samples to be predicted are subjected to Euclidean distance calculation, wherein the calculation formula is as follows:
Dividing M first-order tensor slices with minimum distance of a sample F (N 0-τ1 +1) to be predicted into a reorganization space, and reordering;
M first-order tensor slices with the smallest distance to the sample F (N 0-τ1 +1) to be predicted are reordered from large to small according to the distance;
Where when m=79, the rearranged second order tensor FR is:
further, the second-order tensor FR is subjected to one-way delay embedding transformation to obtain a third-order tensor corresponding to the approximate sequence Comprising the following steps:
obtaining the delay time tau 2 of the embedded transformation and constructing a unique replication matrix corresponding to the second-order tensor FR
According to the linear tensor product and folding operation of Mode-2 of the second-order tensor FR and the unique replication matrix D 21, obtaining a third-order tensor corresponding to the approximate sequence after the second one-way delay embedding transformation is realized
Wherein the third order tensorThe form of (2) is as follows:
further, the pair of third-order tensors Slicing along the time dimension, and decomposing the second-order tensor slice at each moment by using a Tak to obtain a low-rank high-order time sequence/>, which corresponds to the approximate sequenceComprising the following steps:
tensor of third order Slicing along the time sequence dimension according to time to obtain (80-tau 2 +1) second-order tensor slices/>
Slicing the second order tensorD-order differential calculation is carried out to obtain differential second-order tensor slices/>And second order tensor slice set/>
By orthogonal factor matrix setsSet of approximate second order tensor slices/>Projection to approximate core tensor set/>Wherein/>Factor matrix/>Having orthogonal column characteristics, n is equal to the order of the tensor slices, ε is the error value of the Tak decomposition;
Solving the optimization problem of epsilon minimization by adopting an augmented Lagrangian method
Obtaining a second order core tensor set at each moment by Tack decompositionObtaining a low-rank high-order time sequence/>, which corresponds to the approximate sequence, according to the original time sequence arrangement
Further, the sampling two long-short-time memory networks respectively perform low-rank high-order time sequenceAnd low rank higher order time series/>Learning and prediction are carried out, and core tensor/> is obtained through predictionAnd core tensor/>Comprising the following steps:
Low rank high order time series And/>The three dimensions are respectively time delay, similarity or time sequence of higher order, the input length l of the predicted sample is set, namely each sample is/>Dividing output tags, i.e. each tag is/>For low rank high order time series/>And/>Training by using long-short-time memory networks respectively;
For low rank high order time series And/>Is the last sample/>And/>Prediction is carried out by using two long-short-time memory networks respectively to obtain a core tensor/>And core tensor/>
Further, the core tensorAnd core tensor/>Reduction is carried out and spliced to a time sequence/>And time series/>In (c) obtaining tensor/>Sum tensor/>For tensor/>Sum tensor/>Performing a hank's inverse transformation process to obtain a predicted approximate wind speed f (N 0 +1) and a detail wind speed s (N 0 +1), and obtaining a predicted wind speed according to the approximate wind speed f (N 0 +1) and the detail wind speed s (N 0 +1), including:
For the obtained core tensor And/>Using the resulting factor matrix/>For core tensorsAnd/>Performing inverse transformation of Tak decomposition to obtain approximate and detailed second-order tensor slices/>, with high rankAnd/>Arranged into a new tensor/>Sum tensor/>
For tensorSum tensor/>Using an inverse of the hank transform, a first order tensor/>, length N 0 +1, is obtainedAnd/>At this time, approximate wind speed sequence/>And detail wind speed sequence/>The last value/>And/>As a predicted value, the decomposed sequences are added to obtain a predicted wind speed sequence/>The last wind speed value is taken as the predicted wind speed value.
Further, the decomposing the time series obtained by sampling into an approximate sequence and a detail sequence includes:
The time sequence x (t) obtained by sampling is decomposed into an approximate sequence f (t) and a detail sequence s (t) by adopting a singular spectrum analysis method.
The invention adopts another technical scheme that:
An ultra-short term wind speed prediction apparatus, comprising:
At least one processor;
At least one memory for storing at least one program;
The at least one program, when executed by the at least one processor, causes the at least one processor to implement the method described above.
The beneficial effects of the invention are as follows: the invention improves the traditional statistical method for predicting the wind speed, and firstly decomposes the one-dimensional wind speed time sequence in the data processing link to generate two decomposition sequences for prediction, thereby respectively processing the high-frequency characteristic and the low-frequency characteristic; the Hakking is used for embedding two one-dimensional time sequences into a high-dimensional tensor, and expanding the autocorrelation of the time sequences to a higher dimension; the invention is therefore more interpretable than other methods.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following description is made with reference to the accompanying drawings of the embodiments of the present invention or the related technical solutions in the prior art, and it should be understood that the drawings in the following description are only for convenience and clarity of describing some embodiments in the technical solutions of the present invention, and other drawings may be obtained according to these drawings without the need of inventive labor for those skilled in the art.
FIG. 1 is a flowchart illustrating steps of an ultra-short term wind speed prediction method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a time series of wind speeds collected at a reef-climbing weather station in Australia Mi Ermi in accordance with an embodiment of the present invention;
FIG. 3 is a schematic diagram showing the decomposition effect of a singular spectrum analysis method in an embodiment of the present invention;
FIG. 4 is a schematic diagram of the construction of a unique replication matrix D 1 in an embodiment of the present invention;
FIG. 5 is a diagram of a one-way delay embedded transformation process for a first order tensor in an embodiment of the invention;
FIG. 6 is a graph of the Euclidean Distance Rearrangement (EDR) process of the second order tensor of the approximation sequence in an embodiment of the present invention;
FIG. 7 is a diagram of a one-way delay embedded transformation process for a second order tensor in an embodiment of the invention;
FIG. 8 is a graph of a segmentation process for a third-order tensor in an embodiment of the present invention;
FIG. 9 is a schematic diagram of a Take decomposition of a second order tensor slice in an embodiment of the invention;
FIG. 10 is a graph of the results of a comparison of the proposed method of the present invention with SSA-BPNN, SSA-LSSVM, SSA-LSTM prediction methods in an embodiment of the present invention;
FIG. 11 is a graph showing the comparison of the proposed method of the present invention with SSA-tensor ARMA predictions and SSA-tensor ARIMA predictions in an embodiment of the present invention.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the invention. The step numbers in the following embodiments are set for convenience of illustration only, and the order between the steps is not limited in any way, and the execution order of the steps in the embodiments may be adaptively adjusted according to the understanding of those skilled in the art.
In the description of the present invention, it should be understood that references to orientation descriptions such as upper, lower, front, rear, left, right, etc. are based on the orientation or positional relationship shown in the drawings, are merely for convenience of description of the present invention and to simplify the description, and do not indicate or imply that the apparatus or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus should not be construed as limiting the present invention.
In the description of the present invention, a number means one or more, a number means two or more, and greater than, less than, exceeding, etc. are understood to not include the present number, and above, below, within, etc. are understood to include the present number. The description of the first and second is for the purpose of distinguishing between technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or implicitly indicating the precedence of the technical features indicated.
In the description of the present invention, unless explicitly defined otherwise, terms such as arrangement, installation, connection, etc. should be construed broadly and the specific meaning of the terms in the present invention can be reasonably determined by a person skilled in the art in combination with the specific contents of the technical scheme.
As shown in fig. 1, the embodiment provides an ultra-short term wind speed prediction method, which specifically includes the following steps:
s1, acquiring wind speed time data, and sampling the wind speed time data at intervals to obtain a time sequence.
And acquiring the wind speed in the actual scene, wherein the acquisition frequency is 10 minutes, and obtaining a complete original time sequence WS (i). And (3) performing interval sampling on WS (i) according to the step length p predicted in advance, wherein the interval time length is p. Let the maximum number of steps of the sequence be (1+n 0 p), t be the sampling instant, t=1, 2, …, N 0. The time sequence obtained by sampling is as follows:
x(t)={WS(1+p),...,WS(1+tp),…,WS(1+N0p)} (1)
in this example, wind speed data of weather stations in a millmi reef-climbing area of the institute of ocean science in australia was collected, and 52833 data of average wind speed was taken every ten minutes from 1 st in 2013, 1 st, to 31 st 24 th 12 th in 2013. In order to facilitate the visual recognition of the data characteristics, the first 500 wind speed data are drawn into a graph by matlab, the abscissa corresponds to the serial number of each point, the ordinate corresponds to the values of the wind speeds, a Mi Ermi reef climbing wind speed time sequence graph is drawn, and the local data are shown in figure 2. Assuming a sampling interval p=6, the sequence pattern sampled in the first 500 wind speed data is also shown in fig. 2.
S2, decomposing the time sequence obtained by sampling into an approximate sequence and a detail sequence.
In some alternative embodiments, the sampled time series x (t) is decomposed into an approximation series f (t) and a detail series s (t) using a method of Singular Spectrum Analysis (SSA).
In this embodiment, the sequence length of x (t) is N 0, a suitable window length w=12 is selected to perform hysteresis arrangement on the original time sequence to obtain a track matrix, singular value decomposition is performed on the track matrix, the component with the largest contribution is selected as the approximate sequence f (t) according to the size of the singular value, and other components reconstruct the detail sequence s (t). The approximate and detailed sequence patterns obtained after the time series of the original wind speed in fig. 2 are analyzed by singular spectra are shown in fig. 3.
S3, performing Hakking processing on the high-order tensors of the approximate sequence and the detail sequence to obtain a third-order tensor corresponding to the approximate sequenceThird-order tensor/>, corresponding to detail sequence
Wherein, the Hakking process is performed on the high-order tensors of the approximation sequence f (t) and the detail sequence S (t), comprising the following steps S31-S33:
S31, second-order Hakking of the approximate sequence f (t) and the detail sequence S (t).
Regarding the approximate sequences F (t) and S (t) obtained by filtering in the step S2, regarding them as first-order tensors F and S with the size of 1×n 0, respectively, a unique duplication matrix D 1 is constructed according to a selected delay time τ 1, and the constructed example is shown in fig. 4, and then, a one-way delay embedding transformation is performed to change the first-order tensors into second-order tensors F and S, and the transformation process is shown in fig. 5.
As an alternative embodiment, step S31 specifically includes steps S311 to S312:
S311, selecting the delay time tau 1 of the embedding transformation to construct a unique replication matrix D 1.
Regarding the approximate sequences f (t) and S (t) obtained by filtering in the step S2 as first-order tensors f and S with the size of 1×n 0, f= [ f (1), f (2),. The term, f (t), the term, f (N 0)],s=[s(1),s(2),...,s(t),...,s(N0) ], the time t=1, 2, …, N 0, the delay time τ 1 of the embedding transformation is selected, and a unique replication matrix is constructedAs shown in fig. 4, the unique replication matrix D 1 consists of (N 0-τ1 +1) τ 1×τ1 -sized identity matrices spliced:
S312, one-way delay embedding transformation of the first-order tensor.
The one-way delay embedded transform may be expressed as the result of a refolding of the first order tensors f and s with the 1-Mode tensor product ("× 1") of the unique replica matrix D 1, where the result of the linear tensor product-folding operation of Mode-1 of the first order tensors f and D 1 is as follows:
f×1D1=[f(1) f(2) … f(τ1) f(2) f(3) … f(τ1+1) …] (3)
In this embodiment, with τ 1 set to a value of 3, the results of the linear tensor product-fold operation for Mode-1 of the first order tensors f and D 1 can be seen as shown in FIG. 5. The form of the approximate second order tensor F obtained after folding is as follows:
The form of the detail second order tensor S (t) and so on, fig. 5 shows their common graph, with τ 1 =3 for example.
S32, euclidean Distance Rearrangement (EDR) of second-order tensors of the approximate sequence.
Dividing the second-order tensor F along the second dimension to obtain a plurality of first-order tensor slices F (t) and S (t) at different moments, calculating Euclidean distances between the first-order tensor slices F (N 0-τ1 +1) and S (N 0-τ1 +1) to be predicted according to Euclidean distance theory, dividing M first-order tensor slices with the minimum distance into the same sample space, and rearranging the first-order tensor slices in the space into second-order tensors FR according to the sequence of the distances from large to small, wherein the implementation process of the step 3.2) is shown in fig. 6.
As an alternative embodiment, step S32 includes steps S321-S323:
S321, segmentation of second-order tensors.
As can be seen from equation (4), slicing the approximate second-order tensor F along the second dimension at a time gives (N 0-τ1 +1) first-order tensor slices F (t) = [ F (t) F (t+1) … F (t+τ 1)]T,t=1,2,...,N0-τ1 +1).
S322, dividing a sample set.
Based on Euclidean distance theory, F (t) and a sample to be predicted (namely F (N 0-τ1 +1)) at different moments are calculated as follows:
M=79 first order tensor slices with the smallest distance to F (N 0-τ1 +1) are then partitioned into the reassembly space and reordered.
S323, the reordering generates a new approximate second order tensor FR.
The 79 first-order tensor slices with the smallest distance from F (N 0-τ1 +1) are reordered from large to small as { FR (t) }, t=1, 2, …,79, so the rearranged second-order tensor FR is:
s33, performing second one-way delay embedding transformation on the reconstructed approximation and detail two second-order tensors FR and S to obtain approximation and detail two third-order tensors And/>
As an alternative embodiment, step S33 includes S331-S332:
S331, selecting delay time tau 2 of embedding transformation to construct a unique replication matrix D 2.
Selecting delay time tau 2 of embedding transformation, obtaining second-order tensor FR with size tau 1 multiplied by 80 in step S32, and uniquely duplicating matrixThe matrix is formed by splicing (80- τ 2 +1) identity matrixes with the size of τ 2×τ2: the second order tensor S obtained in the step S32 is tau 1×(N0-τ1 +1), and the matrix/>The matrix is formed by splicing [ (N 0-τ1+1)-τ2 +1] tau 2×τ2 -sized unit matrixes, and the form is similar to that shown in a formula (2) and a figure 3.
S332, single-path delay embedded transformation of second-order tensor.
The linear tensor product of Mode-2 of the second-order tensors f and D 1 and the approximate third-order tensor obtained after the folding operationThe form of (2) is as follows:
The form of the third-order tensor S (t) of detail and so on, this process is graphically depicted in fig. 7.
S4, for third-order tensorAnd third order tensor/>Slicing along the time dimension, and decomposing the second-order tensor slice at each moment by using a Tak to obtain a low-rank high-order time sequence/>, which corresponds to the approximate sequenceLow-rank high-order time sequence/>, corresponding to detail sequence
For third order tensorsAnd/>They are considered as J 1×J2 ×T high order time series/>, respectivelyAndSlicing along the time dimension, decomposing the second-order tensor slice at each moment by using a Tack to obtain a low-rank second-order core tensor at each moment, and arranging the low-rank second-order core tensor into two low-rank high-order time sequences/>, namely new approximate time sequences and new detail time sequences according to the original time sequenceAnd/>
As an alternative embodiment, step S4 specifically includes steps S41-S44:
S41, segmentation of third-order tensors.
From equation (5), it is known that the third-order tensor will be approximatedSeen as a higher order time series (time delay, similarity or time delay, respectively) sliced in time along the time series dimension to obtain (80- τ 2 +1) second order tensor slices/>T=1, 2..80- τ 2 +1, this procedure is shown in fig. 8.
The second-order tensor slices are subjected to d-order differential calculation according to the data stability characteristics to obtain differential second-order tensor slicesThe result of the first order difference calculation is as follows:
the result of the high order differential computation and so on, it is suggested to use a second order differential computation for the approximate tensor and a zero order differential computation for the detail tensor.
S42, take decomposition of the second-order tensor slice.
Talcum decomposition through orthogonal factor matrix setsSet of approximate second order tensor slices/>Projection to approximate core tensor set/>A schematic of this process is shown in fig. 9, where n is equal to the order of the tensor slice and epsilon is the error value of the tac decomposition:
Wherein the factor matrix With orthogonal column characteristics, a similar approach can yield a detailed kernel tensor set
S43, minimizing the Take decomposition error.
Solving the epsilon-minimum optimization problem using the augmented Lagrangian method, which is equivalent to the formula:
the optimization problem is solved by using a closed loop solution, i.e. the tensor slices are matrixed along each mode (n), each decision variable is updated in turn, and after a certain iteration, the optimal variable value and the minimum error value are obtained, which means that the factor matrix is updated in turn And approximate core tensor/>The content of step S43 can be understood with reference to the content in the subsequent step S52.
S44, sorting the second-order core tensors into third-order low-rank tensors.
The Talcum decomposition obtains a second-order core tensor set at each momentAnd/>And are arranged into new approximate and detail two three-order low-rank tensors/>, according to the original time sequenceAnd/>
S5, sampling two long-short-time memory networks to respectively carry out low-rank high-order time sequenceAnd low rank higher order time series/>Learning and prediction are carried out, and core tensor/> is obtained through predictionAnd core tensor/>
Will beAnd/>The three dimensions are respectively time delay, similarity or time delay and time sequence of high-order time sequence, pair/>And/>Respectively using two long-short-time memory networks to learn and predict to obtain future core tensorsAnd/>
As an alternative embodiment, step S5 specifically includes steps S51-S53:
s51, training a long-term memory network (LSTM).
Will beAnd/>Can also be regarded as a higher-order time sequence with three dimensions of time delay, similarity or time delay and time sequence respectively, and the input length l of the predicted samples is set, namely each sample is/>Dividing output tags, i.e. each tag is/>Pair/>And/>Training was performed using long and short term memory networks, respectively.
S52, simultaneously minimizing the prediction error and the Take decomposition error.
The model error formula consists of two parts, namely a prediction error and a Take decomposition error:
The predicted value of the memory network is the time long and short. And iteratively updating decision variables by using a closed-loop solution, and updating the approximate core tensor first and then updating the factor matrix. The formula for updating the core tensor is:
Factor matrix The updating method accords with a global optimal solution for solving an orthogonal Procrustes problem maximization problem, so the updating method comprises the following steps:
Wherein the method comprises the steps of And/>Are respectively/>Left and right singular vectors obtained by singular value decomposition of (c).
The long and short term memory network also needs to be retrained one round each time the approximate core tensor and factor matrix are updated one round to reduce the prediction error.
S53, predicting a new core tensor.
For a pair ofAnd/>Is the last sample/>And/>Respectively using two long-short-term memory networks to predict to obtain future core tensor slices/>And/>
S6, tensor of coreAnd core tensor/>Reduction is carried out and spliced to a time sequence/>And time series/>In (c) obtaining tensor/>Sum tensor/>For tensor/>Sum tensor/>Performing inverse hank transformation to obtain a predicted approximate wind speed f (N 0 +1) and a detail wind speed s (N 0 +1), and obtaining a predicted wind speed according to the approximate wind speed f (N 0 +1) and the detail wind speed s (N 0 +1).
For predictionAnd/>Reducing them to/>And/>Splice to time series/>And/>In (3) obtaining a new/>And/>Tensors, and then a series of inverse Hank transforms are used to obtain a predicted approximate wind speed f (N 0 +1) and a detail wind speed s (N 0 +1), which are added to obtain a predicted wind speed WS (t+ (N 0 +1) p). As an alternative embodiment, step 65 specifically includes steps S61-S62:
s61, inverse Take decomposition and transformation.
For newly obtainedAnd/>Using the resulting factor matrix/>Inverse transformation of the Talcum decomposition of them to obtain high-rank approximation and detail second-order tensor slices/>And/>If d-th order difference calculation is used, the tensor slices need to be integrated and restored, and finally the tensor slices are arranged into new/>And/>Tensors.
S62, inverse Hank transformation.
For approximate and detailed third-order tensorsAnd/>Inverse transformation using Hakking yields a first order tensor/>, length N 0 +1And/>The formula of the inverse hank transform is as follows:
in formula (15) Symbolically represented is the Moore-Penrose pseudo-inverse. At this time, approximate wind speed sequence/>And detail wind speed sequence/>The last value/>And/>I.e. predicted value, and the decomposed sequences are added to obtain a predicted wind speed sequence/>The last wind speed value is the predicted wind speed value/>
S7, ultra-short-term wind speed prediction and performance indexes thereof.
Wind speed predictions are classified on a prediction time scale into long-term predictions (in years), mid-term predictions (one week in advance), short-term predictions (4-72 hours in advance) and ultra-short-term predictions (10 minutes-4 hours in advance). The sampling frequency of the wind speed time series of the method of the embodiment is 10 minutes, and the predicted time scale is 60/120/180/240 minutes in advance.
The predicted value needs to be judged by different performance indexes, and the following are several different predicted indexes, and the calculation formula is as follows:
(1) Mean Absolute Error (MAE):
where WS (i) is the true wind speed at time i, Representing the predicted wind speed at time i. The smaller the index, the higher the accuracy of the prediction algorithm, N representing the total number of wind speed data samples for prediction;
(2) Root Mean Square Error (RMSE):
the smaller the index, the closer the predicted data of the prediction algorithm is to the characteristics of the real data.
S71, predicting the wind speed time sequence of the same data set (Australian Mimi reef) by various global prediction methods including an SSA-BPNN method, an SSA-LSSVM (least squares support vector machine) method and an SSA-LSTM method, and comparing the performance indexes of the prediction results of the different methods with those of the method.
Table 1 shows the results of the predicted performance comparisons of the method described in step S71 over a time series of wind speeds. FIG. 10 is a graph of predicted results 60 minutes ahead of the same segment of wind speed sequence for several methods, with only 100 consecutive data selected for comparison for clarity of distinction.
Table 1 performance index (%) -global prediction contrast for wind speed time series predictions
/>
As is apparent from fig. 10 and table 1, the present embodiment proposes a wind speed prediction method that is superior to various classical global prediction methods.
S72, predicting the wind speed time sequence of the same data set (Australian Mimi reef climbing) by using an advanced local prediction method, wherein the advanced local prediction method comprises an SSA-tensor ARMA method and an SSA-tensor ARIMA method, and comparing performance indexes of prediction results of different methods with the method.
Table 2 shows the results of the predicted performance comparisons of the method described in step 7.2) over a time series of wind speeds. FIG. 11 is a graph of predicted results of 60 minutes ahead of the same wind speed sequence for several methods, with only 50 consecutive data selected for comparison for clarity of distinction.
TABLE 2 Performance index (%) -local prediction contrast of wind speed time series prediction results
As is apparent from fig. 11 and table 2, the present embodiment proposes a wind speed prediction method that is superior to an advanced local prediction method.
In summary, compared with the prior art, the method of the embodiment has the following beneficial effects:
(1) Experiments show that the method of the embodiment is superior to some mainstream global prediction methods and most advanced local prediction methods. The method is suitable for ultra-short-term prediction with the sampling frequency of the original wind speed data being 10 minutes and the prediction duration being within four hours in advance. In the embodiment, simulation verification is performed on the established model, so that the prediction result reaches an ideal range, the average absolute error of the prediction result (after normalization) is within 4%, the root mean square error is within 5%, and the accuracy is obviously improved compared with that of a comparison method. Therefore, the method can improve the accuracy of wind speed prediction finally.
(2) The method improves the traditional statistical method for predicting the wind speed, and in the data processing link, the one-dimensional wind speed time sequence is decomposed through singular spectrum analysis to generate two decomposition sequences for prediction, so that the high-frequency characteristic and the low-frequency characteristic can be processed respectively. The haar-gram is used to embed two one-dimensional time sequences into a high-dimensional tensor, extending the autocorrelation of the time sequences to a higher dimension. During the hank's process, fibers/vectors in the approximate tensors are selected and reordered using a similarity search technique, thereby enhancing the translational invariance characteristics in the second-order tensor slices. The invention is therefore more interpretable than other methods.
(3) According to the method, the high-rank tensor is converted into the low-rank tensor through the Talck decomposition, so that information redundancy caused by Halck conversion can be reduced, and compared with other complex high-dimensional prediction models, the calculation cost is remarkably reduced.
The embodiment also provides an ultra-short term wind speed prediction device, which comprises:
At least one processor;
At least one memory for storing at least one program;
The at least one program, when executed by the at least one processor, causes the at least one processor to implement the method as shown in fig. 1.
The ultra-short-term wind speed prediction device provided by the embodiment of the method can be used for executing the method for ultra-short-term wind speed prediction, and any combination of the embodiments of the method can be used for executing the steps, so that the method has the corresponding functions and beneficial effects.
Embodiments of the present application also disclose a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The computer instructions may be read from a computer-readable storage medium by a processor of a computer device, and executed by the processor, to cause the computer device to perform the method shown in fig. 1.
In the foregoing description of the present specification, reference has been made to the terms "one embodiment/example", "another embodiment/example", "certain embodiments/examples", and the like, means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that: many changes, modifications, substitutions and variations may be made to the embodiments without departing from the spirit and principles of the invention, the scope of which is defined by the claims and their equivalents.
While the preferred embodiment of the present application has been described in detail, the present application is not limited to the above embodiments, and various equivalent modifications and substitutions can be made by those skilled in the art without departing from the spirit of the present application, and these equivalent modifications and substitutions are intended to be included in the scope of the present application as defined in the appended claims.
Claims (10)
1. An ultra-short term wind speed prediction method is characterized by comprising the following steps:
acquiring wind speed time data, and sampling the wind speed time data at intervals to acquire a time sequence;
decomposing the time sequence obtained by sampling into an approximate sequence and a detail sequence;
Performing Hakking processing on the high-order tensors of the approximate sequence and the detail sequence to obtain a third-order tensor corresponding to the approximate sequence Third-order tensor/>, corresponding to detail sequence
For third-order tensorAnd third order tensor/>Slicing along the time dimension, and decomposing the second-order tensor slice at each moment by using a Tak to obtain a low-rank high-order time sequence/>, which corresponds to the approximate sequenceLow-rank high-order time sequence/>, corresponding to detail sequence
Sampling two long-short-time memory networks to respectively carry out low-rank high-order time sequenceAnd low rank higher order time series/>Learning and prediction are carried out, and core tensor/> is obtained through predictionAnd core tensor/>
The core tensorAnd core tensor/>Reduction is carried out and spliced to a time sequence/>And time series/>In (c) obtaining tensor/>Sum tensor/>For tensor/>Sum tensor/>Performing inverse hank transformation to obtain a predicted approximate wind speed f (N 0 +1) and a detail wind speed s (N 0 +1), and obtaining a predicted wind speed according to the approximate wind speed f (N 0 +1) and the detail wind speed s (N 0 +1).
2. The ultra-short term wind speed prediction method according to claim 1, wherein the high-order tensors of the approximate sequence and the detail sequence are subjected to a Hakk process to obtain a third-order tensor corresponding to the approximate sequenceThird-order tensor/>, corresponding to detail sequenceComprising the following steps:
Regarding the approximate sequence F (t) and the detail sequence S (t) as a first-order tensor F and a first-order tensor S with the size of 1 XN 0, and performing one-way delay embedding transformation to obtain a second-order tensor F and a second-order tensor S;
dividing a second-order tensor F along a second dimension to obtain a plurality of first-order tensor slices F (t) at different moments, calculating Euclidean distances between the plurality of first-order tensor slices F (t) and a sample to be predicted, dividing M first-order tensor slices with the minimum Euclidean distance into the same sample space, and rearranging the first-order tensor slices in the sample space into second-order tensors FR according to the sequence from the large distance to the small distance;
The second-order tensor FR and the second-order tensor S are subjected to one-way delay embedded transformation to obtain a third-order tensor corresponding to the approximate sequence Third-order tensor/>, corresponding to detail sequence
3. The method according to claim 2, wherein the step of regarding the approximation sequence F (t) as a first-order tensor F with a size of 1×n 0, performing a one-way delay embedding transformation to obtain a second-order tensor F comprises:
Considering the approximation sequence f (t) as a first order tensor f, f= [ f (1), f (2) ], for a1×n 0 size, f (t), f (N 0), time t=1, 2, …, N 0;
Selecting the delay time tau 1 of the embedding transformation to construct a unique replica matrix
And performing single-way delay embedded transformation to obtain a second-order tensor F, wherein the single-way delay embedded transformation is expressed as the result of linear tensor sum-folding operation of the first-order tensor F and the Mode-1 of the unique replication matrix D 1.
4. The ultra-short term wind speed prediction method according to claim 2, wherein the dividing the second order tensor F along the second dimension to obtain a plurality of first order tensor slices F (t) at different moments, calculating the euclidean distance between the plurality of first order tensor slices F (t) and the sample to be predicted, dividing M first order tensor slices with the smallest euclidean distance into the same sample space, and rearranging the first order tensor slices in the sample space into second order tensors FR according to the order from the larger distance to the smaller distance, including:
Slicing the approximate second-order tensor F along the second dimension at time to obtain (N 0-τ1 +1) first-order tensor slices F (t) = [ F (t) F (t+1) … F (t+τ 1)]T,t=1,2,...,N0-τ1 +1;
based on Euclidean distance theory, F (t) at different moments and F (N 0-τ1 +1) samples to be predicted are subjected to Euclidean distance calculation, wherein the calculation formula is as follows:
Dividing M first-order tensor slices with minimum distance of a sample F (N 0-τ1 +1) to be predicted into a reorganization space, and reordering;
M first-order tensor slices with the smallest distance to the sample F (N 0-τ1 +1) to be predicted are reordered from large to small according to the distance;
Where when m=79, the rearranged second order tensor FR is:
5. the ultra-short term wind speed prediction method according to claim 2, wherein the second-order tensor FR is subjected to one-way delay embedding transformation to obtain a third-order tensor corresponding to an approximation sequence Comprising the following steps:
obtaining the delay time tau 2 of the embedded transformation and constructing a unique replication matrix corresponding to the second-order tensor FR
According to the linear tensor product and folding operation of Mode-2 of the second-order tensor FR and the unique replication matrix D 21, obtaining a third-order tensor corresponding to the approximate sequence after the second one-way delay embedding transformation is realized
Wherein the third order tensorThe form of (2) is as follows:
6. the ultra-short term wind speed prediction method according to claim 1, wherein the pair of third-order tensors Slicing along the time dimension, and decomposing the second-order tensor slice at each moment by using a Tak to obtain a low-rank high-order time sequence/>, which corresponds to the approximate sequenceComprising the following steps:
tensor of third order Slicing along the time sequence dimension according to time to obtain (80-tau 2 +1) second-order tensor slices/>
Slicing the second order tensorD-order differential calculation is carried out to obtain differential second-order tensor slices/>And second order tensor slice set/>
By orthogonal factor matrix setsSet of approximate second order tensor slices/>Projection to approximate core tensor set/>Wherein/>Factor matrix/>Having orthogonal column characteristics, n is equal to the order of the tensor slices, ε is the error value of the Tak decomposition;
Solving the optimization problem of epsilon minimization by adopting an augmented Lagrangian method
Obtaining a second order core tensor set at each moment by Tack decompositionObtaining a low-rank high-order time sequence/>, which corresponds to the approximate sequence, according to the original time sequence arrangement
7. The ultra-short term wind speed prediction method according to claim 1, wherein the sampling two long and short term memory networks are respectively used for low-rank high-order time seriesAnd low rank higher order time series/>Learning and prediction are carried out, and core tensor/> is obtained through predictionAnd core tensor/>Comprising the following steps:
Low rank high order time series And/>The three dimensions are respectively time delay, similarity or time sequence of higher order, the input length l of the predicted sample is set, namely each sample is/>Dividing output tags, i.e. each tag is/>For low rank high order time series/>And/>Training by using long-short-time memory networks respectively; for low rank high order time series/>And/>Is the last sample/>And/>Prediction is carried out by using two long-short-time memory networks respectively to obtain a core tensor/>And core tensor/>
8. The ultra-short term wind speed prediction method according to claim 1, wherein the core tensor is calculated based on the calculated core tensorAnd core tensor/>Reduction is carried out and spliced to a time sequence/>And time series/>In (c) obtaining tensor/>Sum tensor/>For tensor/>Sum tensor/>Performing a hank's inverse transformation process to obtain a predicted approximate wind speed f (N 0 +1) and a detail wind speed s (N 0 +1), and obtaining a predicted wind speed according to the approximate wind speed f (N 0 +1) and the detail wind speed s (N 0 +1), including:
For the obtained core tensor And/>Using the resulting factor matrix/>For core tensorsAnd/>Performing inverse transformation of Tak decomposition to obtain approximate and detail second-order tensor slices with high rankAnd/>Arranged into a new tensor/>Sum tensor/>
For tensorSum tensor/>Using an inverse of the hank transform, a first order tensor/>, length N 0 +1, is obtainedAnd/>At this time, approximate wind speed sequence/>And detail wind speed sequence/>The last value/>And/>As a predicted value, the decomposed sequences are added to obtain a predicted wind speed sequence/>The last wind speed value is taken as the predicted wind speed value.
9. The ultra-short term wind speed prediction method according to claim 1, wherein said decomposing said time series of sample acquisitions into an approximate sequence and a detail sequence comprises:
The time sequence x (t) obtained by sampling is decomposed into an approximate sequence f (t) and a detail sequence s (t) by adopting a singular spectrum analysis method.
10. An ultra-short term wind speed prediction apparatus, comprising:
At least one processor;
At least one memory for storing at least one program;
When the at least one program is executed by the at least one processor, the at least one processor is caused to implement the method of any one of claims 1-9.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111370814.0A CN114202107B (en) | 2021-11-18 | 2021-11-18 | Ultra-short-term wind speed prediction method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111370814.0A CN114202107B (en) | 2021-11-18 | 2021-11-18 | Ultra-short-term wind speed prediction method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114202107A CN114202107A (en) | 2022-03-18 |
CN114202107B true CN114202107B (en) | 2024-04-19 |
Family
ID=80648175
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111370814.0A Active CN114202107B (en) | 2021-11-18 | 2021-11-18 | Ultra-short-term wind speed prediction method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114202107B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013037471A (en) * | 2011-08-05 | 2013-02-21 | Nec Corp | Probabilistic model update system, probabilistic model update device, probabilistic model update method, and program |
CN105608271A (en) * | 2015-12-21 | 2016-05-25 | 武汉大学 | Decomposition and optimization based short-term wind speed time series prediction method |
CN112241938A (en) * | 2020-08-21 | 2021-01-19 | 浙江工业大学 | Image restoration method based on smooth Tak decomposition and high-order tensor Hank transformation |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11860682B2 (en) * | 2019-12-23 | 2024-01-02 | Xerox Corporation | Method and system for qualitative reasoning of spatio-temporal physical systems |
-
2021
- 2021-11-18 CN CN202111370814.0A patent/CN114202107B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013037471A (en) * | 2011-08-05 | 2013-02-21 | Nec Corp | Probabilistic model update system, probabilistic model update device, probabilistic model update method, and program |
CN105608271A (en) * | 2015-12-21 | 2016-05-25 | 武汉大学 | Decomposition and optimization based short-term wind speed time series prediction method |
CN112241938A (en) * | 2020-08-21 | 2021-01-19 | 浙江工业大学 | Image restoration method based on smooth Tak decomposition and high-order tensor Hank transformation |
Non-Patent Citations (2)
Title |
---|
三维瞬态风场下覆冰导线舞动响应研究;楼文娟;姜雄;杨伦;;振动与冲击;20161128(第22期);第6-14页 * |
基于张量分解的超光谱图像降秩与压缩;张乐飞;何发智;;武汉大学学报(信息科学版);20170205(第02期);第54-58页 * |
Also Published As
Publication number | Publication date |
---|---|
CN114202107A (en) | 2022-03-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110059878B (en) | Photovoltaic power generation power prediction model based on CNN LSTM and construction method thereof | |
Li et al. | Smart wind speed forecasting approach using various boosting algorithms, big multi-step forecasting strategy | |
EP3161527B1 (en) | Solar power forecasting using mixture of probabilistic principal component analyzers | |
CN110674323B (en) | Unsupervised cross-modal Hash retrieval method and system based on virtual label regression | |
CN110672905A (en) | CNN-based self-supervision voltage sag source identification method | |
CN112634608B (en) | Regional traffic flow prediction method and system | |
CN115829024B (en) | Model training method, device, equipment and storage medium | |
CN109086463B (en) | Question-answering community label recommendation method based on regional convolutional neural network | |
CN114419464A (en) | Twin network change detection model based on deep learning | |
CN112508286A (en) | Short-term load prediction method based on Kmeans-BilSTM-DMD model | |
CN106682782A (en) | Short-term photovoltaic power prediction method based on EWT-KMPMR (empirical wavelet transform and kernel minimax probability machine classification) | |
CN116384576A (en) | Wind speed prediction method, device, system and storage medium | |
CN115423162A (en) | Traffic flow prediction method and device, electronic equipment and storage medium | |
CN114202107B (en) | Ultra-short-term wind speed prediction method and device | |
CN116706907B (en) | Photovoltaic power generation prediction method based on fuzzy reasoning and related equipment | |
CN113837434A (en) | Solar photovoltaic power generation prediction method and device, electronic equipment and storage medium | |
CN116995670A (en) | Photovoltaic power ultra-short-term prediction method based on multi-mode decomposition and multi-branch input | |
CN109299291B (en) | Question-answering community label recommendation method based on convolutional neural network | |
CN117290716A (en) | Photovoltaic array fault diagnosis method based on KPCA-TCN-Attention | |
CN116011655A (en) | Load ultra-short-term prediction method and system based on two-stage intelligent feature engineering | |
Li et al. | An innovated integrated model using singular spectrum analysis and support vector regression optimized by intelligent algorithm for rainfall forecasting | |
CN114630207A (en) | Multi-sensing-node perception data collection method based on noise reduction self-encoder | |
CN113344245A (en) | Hybrid deep learning short-term prediction model, method, storage medium, and computing device | |
CN114237394B (en) | Motion recognition method, device, equipment and medium | |
CN116842855B (en) | Distributed photovoltaic power distribution network output prediction method and device, electronic equipment and medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |