CN115907120A - Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method - Google Patents
Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method Download PDFInfo
- Publication number
- CN115907120A CN115907120A CN202211414113.7A CN202211414113A CN115907120A CN 115907120 A CN115907120 A CN 115907120A CN 202211414113 A CN202211414113 A CN 202211414113A CN 115907120 A CN115907120 A CN 115907120A
- Authority
- CN
- China
- Prior art keywords
- lstm
- wind power
- cnn
- vmd
- formula
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 230000007246 mechanism Effects 0.000 title claims abstract description 39
- 238000004422 calculation algorithm Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 9
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000003068 static effect Effects 0.000 claims description 6
- 238000004804 winding Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000014509 gene expression Effects 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000010248 power generation Methods 0.000 claims description 4
- 230000003190 augmentative effect Effects 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000006073 displacement reaction Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 claims description 2
- 230000010363 phase shift Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 5
- 238000013527 convolutional neural network Methods 0.000 description 13
- 230000006870 function Effects 0.000 description 10
- 238000012549 training Methods 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000003446 memory effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a VMD-CNN-LSTM short-term wind power prediction method based on an attention mechanism, which comprises the following steps of: aiming at the volatility and the non-stability of the wind power signal, the wind power output is decomposed into components with different frequencies through the VMD, each component is predicted by using the CNN-LSTM, and a feature attention mechanism is introduced on the basis of the LSTM to distribute different weights to the features. In order to autonomously identify the incidence relation between the wind power output and each wind power characteristic, a non-traditional method is used to avoid the limitation and selection depending on expert experience incidence rule threshold values, a characteristic attention mechanism is introduced to autonomously calculate the contribution rate of each characteristic quantity, the characteristic weight is corrected, and the stability and the accuracy of the sequence prediction effect are effectively improved.
Description
Technical Field
The invention relates to the technical field of new energy power generation and smart power grids, in particular to a VMD-CNN-LSTM short-term wind power prediction method based on an attention mechanism.
Background
Wind energy has gained increasing popularity in modern power systems as a clean and economical source of electricity generation. However, due to the randomness, volatility and intermittency of wind energy itself, large-scale integration of wind energy introduces significant uncertainties and risks to the safe operation and wind energy utilization of the power supply side of the power system, which makes the operation of the system more challenging.
In order to reduce the adverse effect of uncertainty of wind power on the operation of the power system, a series of researches have been carried out by more scholars to improve the prediction accuracy of wind power. The prediction method mainly comprises a traditional statistical method and a prediction method based on machine learning. The statistical methods mainly include Multiple Linear Regression (MLR), time series analysis, kalman filtering and the like, and the principles and modeling of the methods are simple, but the prediction effect is not obvious when the data sample capacity is large. Another class of methods is based on machine learning algorithms, such as gray systems, artificial neural networks, support Vector Machines (SVMs), and Gaussian Processes (GPs). Among them, back Propagation (BP) neural networks and SVMs are most widely used. However, the above method lacks consideration of time series correlation, and has a problem of being unable to converge efficiently when there are many training samples.
In recent years, a deep learning algorithm represented by a long-short term memory (LSTM) network is better applied to the field of short-term wind power prediction, the LSTM network can fully mine the internal correlation among time series data, and when the LSTM network is characterized by discontinuous data, the prediction accuracy is not high. The CNN network can fully extract features, but the inherent connectivity among data is not strong, and wind power has non-stationarity and strong interference. Therefore, data characteristics can be deeply mined by using the CNN-LSTM network, the wind power sequence is preprocessed by combining the VMD algorithm, the anti-interference performance of the wind power sequence is improved, an attention mechanism is added in the training process, different weights are adaptively distributed to different characteristics, the weight coefficient of key characteristics is improved, the complexity of the characteristics in the CNN-LSTM network model training is reduced, and the prediction precision is improved.
Patent CN113642225A discloses a CNN-LSTM short-term wind power prediction method based on attention mechanism, which has good performance in data prediction, but because the wind power sequence has the characteristics of non-stationarity and strong noise, further optimization in data processing is needed to improve the anti-interference capability of the wind power sequence itself.
Disclosure of Invention
The invention aims to disclose a VMD-CNN-LSTM short-term wind power prediction method based on an attention mechanism, and the method is invented on the basis of a patent CN113642225A, and realizes the decomposition of a wind power sequence by constructing and solving a constraint variation problem in the aspect of data processing, so that the problems of modal aliasing, overcladding, undercladding, boundary effect and the like are effectively avoided, and the method has the advantages of better complex data decomposition precision, better noise interference resistance and the like, and further improves the prediction precision of the wind power sequence.
In order to achieve the purpose, the invention provides a VMD-CNN-LSTM short-term wind power prediction method based on an attention mechanism, which comprises the following steps:
s1, collecting historical power generation data of a wind power plant and historical meteorological data of a region, cleaning power data, standardizing meteorological characteristic quantities and bearing characteristic quantities at different moments, and performing Ordinal-Encoder characteristic coding on non-digital characteristics to complete data preprocessing;
s2, decomposing the historical wind power time sequence data by adopting a VMD algorithm to obtain K modal components { IMF [1], IMF [2], IMF [3], \8230, IMF [ K ] } with different central frequencies;
the decomposition steps are as follows:
s21, calculating a single-side frequency spectrum of each modal component based on a Hilbert transform method;
s22, carrying out treatment on each modal component
So that it is phase-shifted to the center frequency of the mode itself;
s23, estimating the estimation bandwidth of each sub-signal of the modal component according to the Gaussian smoothness of the frequency shift signal, and enabling the estimation bandwidth sum of each sub-signal to be minimum; assuming that an original signal is decomposed into k components, the decomposition sequence is guaranteed to be a modal component with limited bandwidth and a center frequency, and the variation problem with constraint conditions is constructed as shown in the formulas (1) and (2):
original signal:
modeling equation:
in the formula: u. u k In the form of a modal signal,
as partial differential function over time t, f is the time series, k is the modal number, a convolution operator, w k Is the center frequency of the kth mode, delta (t) is a unit impulse function, j is the coefficient of the imaginary part of Hibolter, and->
Is a least square method;
s24, transforming the variation problem into a non-constraint problem through a Lagrange multiplier lambda, solving the optimal solution of a variation constraint model, and obtaining an augmented Lagrange expression as shown in a formula (3):
in the formula: alpha is a secondary penalty factor, and lambda is a Lagrange multiplier, so thatUsing alternative direction multiplier method to calculate and obtain u k And w k Simultaneously searching a saddle point of the formula (3), which is the optimal solution in the formula (1), and obtaining the saddle point through alternate updating
The expressions (A) are as follows:
in the formula: epsilon is a convergence criterion, N is the maximum iteration number, and gamma is a displacement vector; (ii) a
S3, establishing a VMD-CNN-LSTM prediction model aiming at the k decomposed components, adding an attention mechanism to mine the correlation between the wind power output and the related meteorological characteristic quantity and the bearing characteristic quantity after data preprocessing, wherein hyper-parameters of the attention mechanism-based VMD-CNN-LSTM prediction model are obtained by optimizing a hyper-opt Bayes algorithm, and accordingly establishing a corresponding final prediction model.
As a further improvement of the present invention, in step S3, CNN is a one-dimensional convolution network for extracting static features of the input vector, and its output is:
in the formula, X t Input feature vector for time t；w j Is a convolution kernel weight matrix; b is a deviation value; and ke is the number of convolution kernels.
As a further improvement of the present invention, in step S3, the LSTM network specifically includes an LSTM hidden layer, an attention layer, a Dropout layer, and a full connection layer, where the process of extracting features by the LSTM hidden layer is as follows:
i t ＝σ(W ix X t +W ih h t-1 +b i )
f t ＝σ(W fx X t +W fh h t-1 +b f )
O t ＝σ(W Ox X t +W Oh h r-1 +b o )
in the formula i t Is an input gate; f. of t To forget the door; o t Is an output gate;
is the current neuron candidate value, W ix ，W ih ，W fx ，W fh ，W Οx ，W Οh Static features X representing the corresponding gate and the current input, respectively t And the last unit output h t-1 Multiplying to obtain a matrix weight; b is a mixture of i ，b f ，b O Is a bias term; sigma is sigmoid function;
new state value C t Then, the output value h of the hidden layer is obtained t The calculation formula is as follows:
h t ＝O t ×tanh(C t ) (10)
wherein x represents a bitwise product of elements in each multiplication vector;
calculating the feature vector extracted by the hidden layer by using an attention layer to obtain a feature weight, and multiplying the feature weight by the feature vector extracted by the hidden layer;
the multiplication process is as follows: calculating attention weight a distributed to elements in the last layer of output sequence of the LSTM hidden layer at the current time t ti The formula is as follows:
where i denotes the sequence number of an element in the output sequence of the LSTM hidden layer, T h Indicating the length of the output sequence of the LSTM hidden layer, e ti Representing the matching degree between the element to be coded and other elements in the output sequence of the LSTM hidden layer; calculating a feature weight vector, wherein the formula is as follows:
a′ t，i ＝H(D，C t ，h)
in the formula, H represents a feature weight vector function, H t An output sequence representing the t moment of the LSTM hidden layer, h an output characteristic sequence representing all the moments of the last layer of the LSTM hidden layer, C t Representing LSTM hidden layer output sequence h t Hidden layer state of corresponding attention mechanism; a' t,i Denotes h t The weight of the ith element;
and wind power prediction is realized through VMD-CNN-LSTM model output based on an attention mechanism.
As a further improvement of the present invention, the meteorological characteristic quantity in step S1 includes an ambient temperature, and the bearing characteristic quantity includes: blade pitch angle, control box temperature, gearbox bearing temperature, gearbox oil temperature, generator speed, generator winding temperature, hub temperature, main box temperature, nacelle position, reaction power, and rotor speed.
Compared with the prior art, the invention has the beneficial effects that:
(1) The invention provides a attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method, which designs a corresponding wind power prediction model, wherein the model is formed by combining a CNN-LSTM network and a VMD algorithm; the CNN is not completely used and a time sequence is learned, so a method of combining the CNN-LSTM network is adopted, and a certain memory effect is achieved; aiming at the defect of low training speed of an LSTM network, a local feature and extraction module is added, and a convolutional neural network structure is used for pre-extracting the local feature; in order to reduce modal aliasing and noise interference, a VMD algorithm is used for decomposing the wind power into a plurality of single-component amplitude modulation and frequency modulation signals, so that the problems of end effects and false components in the iteration process are avoided; meanwhile, the attention mechanism can distribute different weights to hidden layers in the VMD-CNN-LSTM, highlight key features in training, improve prediction accuracy and is an ideal model for processing long-term time sequence data problems.
(2) The attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method provided by the invention is combined with the hpyeropt algorithm to realize the optimal solution of the parameters.
Drawings
FIG. 1 is a schematic flow diagram of a VMD-CNN-LSTM short-term wind power prediction method based on an attention mechanism;
FIG. 2 is a VMD-CNN-LSTM model in the VMD-CNN-LSTM short-term wind power prediction method based on the attention mechanism of the present invention;
FIG. 3 is an LSTM network structure in the VMD-CNN-LSTM short-term wind power prediction method based on attention mechanism of the present invention;
FIG. 4 is a CNN network structure in the VMD-CNN-LSTM short-term wind power prediction method based on attention mechanism of the present invention;
FIG. 5 is wind power forecast in a VMD-CNN-LSTM short-term wind power forecast method based on an attention mechanism according to the present invention;
FIG. 6 is an IMF component spectrum analysis in the attention-based VMD-CNN-LSTM short-term wind power prediction method of the present invention;
FIG. 7 shows a central mode of the attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method.
Detailed Description
The present invention is described in detail with reference to the embodiments shown in the drawings, but it should be understood that these embodiments are not intended to limit the present invention, and those skilled in the art should understand that functional, methodological, or structural equivalents or substitutions made by these embodiments are within the scope of the present invention.
Please refer to fig. 1 to 7, which illustrate an embodiment of a method for predicting short-term wind power based on attention mechanism of VMD-CNN-LSTM according to the present invention.
A short-term wind power prediction method of a VMD-CNN-LSTM based on an attention mechanism is characterized in that an optimized solution of parameters is realized by combining an hpyeropt Bayes algorithm based on a prediction model of the VMD-CNN-LSTM based on the attention mechanism, and the prediction accuracy is improved. The method specifically comprises the following steps: s1, collecting historical power generation data of a wind power plant and historical meteorological data of a region, cleaning power data, standardizing meteorological characteristic quantities and bearing characteristic quantities at different moments, and performing Ordinal-Encoder characteristic coding on non-digital characteristics to complete data preprocessing; the meteorological characteristic quantity includes an ambient temperature, and the bearing characteristic quantity includes: blade pitch angle, control box temperature, gearbox bearing temperature, gearbox oil temperature, generator speed, generator winding temperature, hub temperature, main box temperature, nacelle position, reaction power, and rotor speed. The method comprises the following specific steps: 1) And selecting the wind power data from 1/2017 to 3/2020/20/118224 data. And decomposing the original wind power data sequence into a series of modal components with different characteristics by using a VMD technology. Assuming each mode is a finite bandwidth with a center frequency, the variational problem is described as seeking k mode functions u k (t) minimizing the sum of the estimated bandwidths of each mode, with the constraint that the sum of the modes is equal to the input signal f, the specific steps are as follows:
s2, decomposing the historical wind power time sequence data by adopting a VMD algorithm to obtain K modal components { IMF [1], IMF [2], IMF [3], \8230, IMF [ K ] } with different central frequencies;
the decomposition steps are as follows:
s21, calculating a single-side frequency spectrum of each modal component based on a Hilbert transform method;
s22, carrying out treatment on each modal component
So that it is phase-shifted to the center frequency of the mode itself;
s23, estimating the estimation bandwidth of each sub-signal of the modal component according to the Gaussian smoothness of the frequency shift signal, and enabling the estimation bandwidth sum of each sub-signal to be minimum; assuming that an original signal is decomposed into k components, the decomposition sequence is guaranteed to be a modal component with limited bandwidth and a center frequency, and the variation problem with constraint conditions is constructed as shown in the formulas (1) and (2):
original signal:
modeling equation:
in the formula: u. of k Is a modal signal, f is a time series, k is a modal number, w is a convolution operator k Is the center frequency of the kth mode, delta (t) is a unit impulse function, j is a Hilbert imaginary signal coefficient,
is a least square method;
s24, transforming the variation problem into a non-constraint problem through a Lagrange multiplier lambda, solving the optimal solution of a variation constraint model, and obtaining an augmented Lagrange expression as shown in a formula (3):
in the formula: alpha is a secondary penalty factor, lambda is a Lagrange multiplier, and u is calculated by using an alternative direction multiplier method k And w k Simultaneously searching a saddle point of the formula (3), which is the optimal solution in the formula (1), and obtaining the saddle point through alternate updating
The expressions (A) are as follows:
in the formula: epsilon is a convergence criterion, N is the maximum iteration number, and gamma is a displacement vector;
in the process of collecting wind power data, the conditions of data loss and data error can occur, the data needs to be preprocessed, and the processing steps are as follows:
and S11, the wind power data is sequential and generally continuous and smooth, abnormal data is found out by checking the stationarity of the data, and the abnormal data is processed by adopting a method of filling values at the same time in the previous day according to the similarity of the values at the same time every day.
S12, for the missing value of the wind power original data, the method adopts a K-means clustering mode to take three days with the maximum similarity, and then the wind power value at the required moment is filled with the average value of the corresponding moment of the three days. The characteristics of the cluster are other characteristics than the wind-out power, such as ambient temperature, bearing shaft temperature, blade 1, 2, 3 pitch angle, control box temperature, gearbox bearing temperature, gearbox oil temperature, generator speed, generator winding 1, 2 temperature, hub temperature, main box temperature, nacelle position, reaction power, rotor speed, turbine state, wind generator model, wind direction, wind speed, etc.
S13, constructing an input feature vector of the VMD-CNN-LSTM model, and selecting feature factors including the ambient temperature A at the moment of predicting points t Bearing shaft temperature B t Blade 1, 2, 3 pitch angle B1P t 、B2P t 、B3P t Temperature GB of gear box bearing t And oil temperature GO of gear box t And the rotational speed GR of the generator t Temperature GW1 of generator winding 1, 2 t 、GW2 t Temperature H of hub t Main box temperature MB t Nacelle position NP t Reaction power RP t And the rotor speed RRP t Turbine state TS t Model G of wind driven generator t Wind direction WD t Wind speed WS t (ii) a The final input feature vector is denoted as X t ＝[A t ，B t ，B1P t ，B2P t ，…WS t ]。
S3, establishing a VMD-CNN-LSTM prediction model aiming at the k decomposed components, adding an attention mechanism to mine the correlation between the wind power output and the related meteorological characteristic quantity and the bearing characteristic quantity after data preprocessing, wherein hyper-parameters of the attention mechanism-based VMD-CNN-LSTM prediction model are obtained by optimizing a hyper-opt Bayes algorithm, and accordingly establishing a corresponding final prediction model.
S31, inputting the preprocessed feature vector into a CNN network for training, extracting static features of the input vector through operations such as convolution, pooling and the like of the CNN network, inputting the features extracted by the CNN network into a hidden layer of an LSTM network for training, and obtaining the trained feature vector, wherein the specific steps are as follows: s31, pre-extracting local features by adopting a CNN structure;
among many neural network structures, CNN is commonly used for implementing image recognition, image classification, target detection and face recognition, and the structure has neurons with learnable weights and bias values, which can increase low-level features of data, combine the low-level features into multi-level features as the network deepens, and guide subsequent models to learn and adjust such features, and the specific structure thereof is shown in fig. 3. The CNN structure makes the forward transfer function more efficient and reduces the number of parameters in the network by compiling specific features into a convolutional structure. The CNN network structure can realize the characteristic extraction of a time axis, and the output is as follows:
in the formula, X t The input feature vector at the time t; w is a j Is a convolution kernel weight matrix; b is a deviation value; ke is the number of convolution kernels.
S32, wind power prediction is carried out by using an LSTM model;
the LSTM has good memory capacity, and solves the problems of gradient disappearance and gradient explosion of a Recurrent Neural Network (RNN). The method can learn the long-time-sequence dependence information, and can search the rule information from the wind power historical data when the wind power is predicted.
The LSTM network comprises a basic LSTM layer, an attribution layer, a Dropout layer and a last Dense layer, firstly, the features extracted by the CNN network are input into the LSTM layer for feature extraction, wherein the LSTM hidden layer mainly has the function of extracting valuable information through continuous learning and training and forgetting to lose worthless information. The specific structure is shown in fig. 3. Three new 'gates' are added compared with the common RNN, and the three new 'gates' are respectively an input gate i t Forgetting door f t O-shaped output gate t And the current unit candidate value
The calculation formula is as follows:
i t ＝σ(W ix X t +W ih h t-1 +b i )
f t ＝σ(W fx X t +W fh h t-1 +b f )
O t ＝σ(W Ox X t +W Oh h t-1 +b o )
in the formula i t Is an input gate; f. of t To forget the door; o t Is an output gate;
is the current neuron candidate value, W ix ，W ih ，W fx ，W fh ，W Οx ，W Οh Static features X representing the corresponding gate and the current input, respectively t And the last unit output h t-1 Matrix weights obtained by multiplication; b is a mixture of i ，b f ，b Ο Is a bias term; sigma is sigmoid function;
new state value C t Then, the output value h of the hidden layer is obtained t The calculation formula is as follows:
h t ＝O t ×tanh(C t ) (10)
wherein x represents a bitwise product of elements in each multiplication vector;
calculating the feature vector extracted by the hidden layer by utilizing an attention layer to obtain a feature weight, and multiplying the feature weight by the feature vector extracted by the hidden layer;
the multiplication process is as follows: calculating attention weight a distributed to elements in the last layer of output sequence of the LSTM hidden layer at the current time t ti The formula is as follows:
where i denotes the sequence number of an element in the output sequence of the LSTM hidden layer, T h Indicating the length of the output sequence of the LSTM hidden layer, e ti Indicating the degree of match between the element to be encoded and other elements in the LSTM hidden layer output sequence. Calculating a feature weight vector, wherein the formula is as follows:
a t，i ＝H(D，C t ，h)
wherein H represents a feature weight vector function, H t An output sequence representing the t moment of the LSTM hidden layer, h an output characteristic sequence representing all the moments of the last layer of the LSTM hidden layer, C t Representing LSTM hidden layer output sequence h t Hidden layer state of the corresponding attention mechanism. a' t,i Represents h t The weight of the ith element;
and wind power prediction is realized through VMD-CNN-LSTM model output based on an attention mechanism.
And (3) training the VMD-CNN-LSTM network including the attention mechanism by using the data acquired in the step (S1) and adopting a hyper-pt Bayesian algorithm, and realizing the optimal solution of parameters so as to obtain a trained wind power prediction model.
In step S3, after the model is trained, the validation set is used to perform validation, and finally the test set is predicted, wherein the accuracy of the model can be evaluated by mean absolute percentage error MAPE, and the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
to predict value, y i Are true values.
In summary, according to the attention mechanism-based VMD-CNN-LSTM short-term wind power prediction model provided by the invention, various factors such as environment temperature, bearing shaft temperature, blade 1, 2 and 3 pitch angles, control box temperature, gearbox bearing temperature, gearbox oil temperature, generator rotating speed, generator winding 1 and 2 temperature, hub temperature, main box temperature, cabin position, reaction power, rotor rotating speed, turbine state, wind generator model, wind direction, wind speed and the like are considered, local characteristics and time sequence characteristics of wind power data can be effectively extracted, different weights are distributed, data noise is reduced, and accurate prediction of short-term wind power is realized.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.
Claims (4)
1. A VMD-CNN-LSTM short-term wind power prediction method based on an attention mechanism is characterized by comprising the following steps:
s1, collecting historical power generation data of a wind power plant and historical meteorological data of a region, cleaning power data, standardizing meteorological characteristic quantities and bearing characteristic quantities at different moments, and performing Ordinal-Encoder characteristic coding on non-digital characteristics to complete data preprocessing;
s2, decomposing the historical wind power time sequence data by adopting a VMD algorithm to obtain K modal components { IMF [1], IMF [2], IMF [3], \8230, IMF [ K ] } with different central frequencies;
the decomposition steps are as follows:
s21, calculating a single-side frequency spectrum of each modal component based on a Hilbert transform method;
s22, carrying out treatment on each modal component
The index correction of (1) to phase shift it to the center frequency of the mode itself;
s23, estimating the estimation bandwidth of each sub-signal of the modal component according to the Gaussian smoothness of the frequency shift signal, and enabling the estimation bandwidth sum of each sub-signal to be minimum; assuming that an original signal is decomposed into k components, the decomposition sequence is guaranteed to be a modal component with limited bandwidth and a center frequency, and the variation problem with constraint conditions is constructed as shown in the formulas (1) and (2):
original signal:
modeling equation:
in the formula: u. u k For modal signals, f is the time series, k is the number of modes, the convolution operator, w k Is the center frequency of the kth mode, delta (t) is the unit impulse function, j is the coefficient of the Hibolter imaginary signal,
is a least square method;
s24, transforming the variation problem into a non-constraint problem through a Lagrange multiplier lambda, solving the optimal solution of a variation constraint model, and obtaining an augmented Lagrange expression as shown in a formula (3):
in the formula: alpha is a secondary penalty factor, lambda is a Lagrange multiplier, and u is calculated by using an alternative direction multiplier method k And w k Simultaneously searching the 'saddle point' of the formula (3), which is the optimal solution in the formula (1), and alternately updatingThen obtain
Are as follows:
in the formula: epsilon is a convergence criterion, N is the maximum iteration number, and gamma is a displacement vector;
s3, establishing a VMD-CNN-LSTM prediction model aiming at the k decomposed components, adding an attention mechanism to mine the correlation between the wind power output and the related meteorological characteristic quantity and the bearing characteristic quantity after data preprocessing, wherein hyper-parameters of the attention mechanism-based VMD-CNN-LSTM prediction model are obtained by optimizing a hyper-opt Bayes algorithm, and accordingly establishing a corresponding final prediction model.
2. The attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method of claim 1, wherein: in step S3, CNN is a one-dimensional convolution network for extracting static features of the input vector, and its output is:
in the formula, X t The input feature vector at the time t; w is a j Is a convolution kernel weight matrix; b is a deviation value; ke is the number of convolution kernels.
3. The attention-based VMD-CNN-LSTM short-term wind power prediction method of claim 1, characterized in that: in step S3, the LSTM network specifically includes an LSTM hidden layer, an attention layer, a Dropout layer, and a full connection layer, where the process of extracting features from the LSTM hidden layer is as follows:
i t ＝σ(W ix X t +W ih h t-1 +b i )
f t ＝σ(W fx X t +W fh h t-1 +b f )
O t ＝σ(W Ox X t +W Oh h r-1 +b o )
in the formula i t Is an input gate; f. of t To forget the door; o < O > of a compound t Is an output gate;
for the current neuron candidate value, W ix ，W ih ，W fx ，W fh ，W Οx ，W Οh Static features X representing the corresponding gate and the current input, respectively t And the last unit output h t-1 Matrix weights obtained by multiplication; b i ，b f ，b Ο Is a bias term; sigma is a sigmoid function;
new state value C t Then, the output value h of the hidden layer is obtained t The calculation formula is as follows:
h t ＝O t ×tanh(C t ) (10)
wherein x represents a bitwise product of elements in each multiplication vector;
calculating the feature vector extracted by the hidden layer by using an attention layer to obtain a feature weight, and multiplying the feature weight by the feature vector extracted by the hidden layer;
the multiplication process is as follows: calculating the attention weight a distributed to the elements in the last layer output sequence of the LSTM hidden layer by the current time t ti The formula is as follows:
where i denotes the sequence number of an element in the output sequence of the LSTM hidden layer, T h Indicating the length of the output sequence of the LSTM hidden layer, e ti Representing the matching degree between the element to be coded and other elements in the LSTM hidden layer output sequence; calculating a feature weight vector, wherein the formula is as follows:
a’ t，j ＝H(D，C t ，h)
in the formula, H represents a characteristic weight vector function, H represents an output characteristic sequence of all moments of the last layer of an LSTM hidden layer, and C t Representing LSTM hidden layer output sequence h t Hidden layer state of corresponding attention mechanism, h t Represents the output sequence at time t of the LSTM hidden layer, a' t,i Represents h t The weight of the ith element;
and wind power prediction is realized through VMD-CNN-LSTM model output based on an attention mechanism.
4. The attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method of claim 1, wherein: in step S1, the meteorological characteristic quantities include an ambient temperature, and the bearing characteristic quantities include: blade pitch angle, control box temperature, gearbox bearing temperature, gearbox oil temperature, generator speed, generator winding temperature, hub temperature, main box temperature, nacelle position, reaction power, and rotor speed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211414113.7A CN115907120A (en) | 2022-11-11 | 2022-11-11 | Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211414113.7A CN115907120A (en) | 2022-11-11 | 2022-11-11 | Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115907120A true CN115907120A (en) | 2023-04-04 |
Family
ID=86493278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211414113.7A Withdrawn CN115907120A (en) | 2022-11-11 | 2022-11-11 | Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115907120A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117892113A (en) * | 2024-03-13 | 2024-04-16 | 广东工业大学 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
-
2022
- 2022-11-11 CN CN202211414113.7A patent/CN115907120A/en not_active Withdrawn
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117892113A (en) * | 2024-03-13 | 2024-04-16 | 广东工业大学 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
| CN117892113B (en) * | 2024-03-13 | 2024-06-11 | 广东工业大学 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860982B (en) | VMD-FCM-GRU-based wind power plant short-term wind power prediction method | |
| Liang et al. | A novel wind speed prediction strategy based on Bi-LSTM, MOOFADA and transfer learning for centralized control centers | |
| CN108197648B (en) | Hydroelectric generating set fault diagnosis method and system based on LSTM deep learning model | |
| CN111815033B (en) | Marine wind power prediction method based on RCNN and meteorological time sequence characteristics | |
| Cheng et al. | A novel deep class-imbalanced semisupervised model for wind turbine blade icing detection | |
| CN109146162B (en) | A kind of probability wind speed forecasting method based on integrated Recognition with Recurrent Neural Network | |
| CN109242212A (en) | A kind of wind-powered electricity generation prediction technique based on change Mode Decomposition and length memory network | |
| CN111401755B (en) | Multi-new-energy output scene generation method, device and system based on Markov chain | |
| CN113642225A (en) | CNN-LSTM short-term wind power prediction method based on attention mechanism | |
| CN114757427B (en) | Autoregressive-corrected LSTM intelligent wind power plant ultra-short-term power prediction method | |
| Liu et al. | A fault diagnosis intelligent algorithm based on improved BP neural network | |
| CN114219181B (en) | Wind power probability prediction method based on transfer learning | |
| Zheng et al. | Generative probabilistic wind speed forecasting: A variational recurrent autoencoder based method | |
| Araya et al. | A Multi-Scale Model based on the Long Short-Term Memory for day ahead hourly wind speed forecasting | |
| CN115204035A (en) | Generator set operation parameter prediction method and device based on multi-scale time sequence data fusion model and storage medium | |
| Wei et al. | An ensemble multi-step forecasting model for ship roll motion under different external conditions: A case study on the South China Sea | |
| Ehsan et al. | Wind speed prediction and visualization using long short-term memory networks (LSTM) | |
| CN117498319A (en) | Distributed photovoltaic short-term power prediction method based on fusion clustering and VQC-LSTM | |
| CN115907120A (en) | Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method | |
| CN114596726A (en) | Parking position prediction method based on interpretable space-time attention mechanism | |
| CN116702831A (en) | Hybrid short-term wind power prediction method considering massive loss of data | |
| CN118134284A (en) | Deep learning wind power prediction method based on multi-stage attention mechanism | |
| Tuyen et al. | A combination of novel hybrid deep learning model and quantile regression for short‐term deterministic and probabilistic PV maximum power forecasting | |
| CN114386666A (en) | Wind power plant short-term wind speed prediction method based on space-time correlation | |
| CN116960975A (en) | Photovoltaic power generation amount prediction method and device |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20230404 |
|
| WW01 | Invention patent application withdrawn after publication |