CN115293415A - Multi-wind-farm short-term power prediction method considering time evolution and space correlation - Google Patents
Multi-wind-farm short-term power prediction method considering time evolution and space correlation Download PDFInfo
- Publication number
- CN115293415A CN115293415A CN202210900738.8A CN202210900738A CN115293415A CN 115293415 A CN115293415 A CN 115293415A CN 202210900738 A CN202210900738 A CN 202210900738A CN 115293415 A CN115293415 A CN 115293415A
- Authority
- CN
- China
- Prior art keywords
- wind power
- wind
- data
- time
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000004927 fusion Effects 0.000 claims abstract description 25
- 230000007246 mechanism Effects 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 33
- 239000013598 vector Substances 0.000 claims description 26
- 238000010606 normalization Methods 0.000 claims description 23
- 238000012549 training Methods 0.000 claims description 22
- 238000004364 calculation method Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 12
- 230000000737 periodic effect Effects 0.000 claims description 11
- 238000012795 verification Methods 0.000 claims description 11
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000005259 measurement Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000010248 power generation Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims description 2
- 238000000354 decomposition reaction Methods 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 abstract description 3
- 230000007547 defect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 15
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 241000288105 Grus Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Primary Health Care (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (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 multi-wind-farm short-term power prediction method considering time evolution and space correlation, which is mainly divided into four modules: the input module collects and preprocesses data, and the target is historical power and meteorological prediction data of a multi-wind farm in a target area; the time evolution mode tracking module respectively extracts a time sequence and a multi-periodicity time evolution mode of historical wind power data through the gated circulation unit and the multi-core convolution layer; the spatial correlation mode attention module introduces a time-varying mode attention mechanism to endow correlation weights to different time evolution modes of the multi-spatial variable; and finally, the output module outputs the day-ahead prediction scene of the power of the multiple wind power plants. According to the method, the time evolution mode of the wind power sequence is completely considered by constructing the time-space fusion multi-wind-farm short-term power prediction model with deep learning capability, the defect that the space dependency relationship is extracted statically by various existing models is overcome, and the purposes of improving prediction precision and robustness are achieved.
Description
Technical Field
The invention belongs to the field of renewable energy power generation and comprehensive consumption, and particularly relates to a multi-wind-farm short-term power prediction method considering time evolution and space correlation.
Background
In recent years, the installed wind power capacity and the proportion of the installed wind power capacity on the power generation side in China are continuously increased, the wind power consumption and the wind abandon problem in large-scale wind power integration are increasingly highlighted due to the natural strong randomness and uncertainty of wind power. The accurate short-term wind power prediction can effectively deal with the inherent randomness and volatility of the wind power, and provides basis for the dispatching operation and the optimization decision of the power system, so that the safe and economic operation after large-scale wind power integration is ensured. Therefore, the research on the short-term wind power prediction technology has important practical significance.
At present, the research on wind power prediction is more and more intensive at home and abroad, and a prediction method system is continuously perfected and expanded. Compared with a single wind power plant, the power prediction of multiple wind power plants has the challenges of multidimensional input variables, complex space-time relation and mutual interleaving. Because the meteorological evolution process has space-time continuity, the wind speed or wind power between wind power plants in the region is closely related. The multi-wind power plant not only has a plurality of different time evolution modes, but also has spatial correlation among the different time evolution modes. The spatial and temporal correlation of the power of the multiple wind power plants is utilized to realize information sharing reciprocity, and the prediction precision and efficiency of the power of the multiple wind power plants can be effectively improved.
In the prior art, a deep learning method applied to multi-wind farm short-term power prediction mainly includes: convolutional Neural Network (CNN), recurrent Neural Network (RNN) and their variants, can be used for extraction of the temporal or spatial correlation of wind-power sequences. For example, an ultrashort-term wind power prediction method based on data driving and deep learning discloses: and extracting the space-time characteristics of the wind power plants by adopting a multi-scale convolution neural network and a GRU (generalized regression Unit), and mining the space-time correlation between adjacent wind power plants. However, in the short-term wind power prediction problem, factors influencing the prediction accuracy not only include the time sequence and spatial correlation of historical wind power data, but also include the potential periodic characteristics of wind power in the time evolution process, and under the combined action of wind processes with different time scales, the wind power sequence may contain periods with various time scales, but the prior art documents only consider the multi-periodicity of wind power places, and do not extract the periodic characteristics or only extract the single-period characteristics; in addition, most of the existing wind power prediction models including the model are static extraction of multi-wind-power-plant space correlation, and the difference and the dynamics of the multi-wind-power-plant power time evolution mode cannot be deeply excavated.
Disclosure of Invention
The method aims to introduce GRUs and multi-core convolution layers, and deeply excavate time sequence information and potential multi-period characteristics of a multi-wind power plant so as to extract a complete time evolution mode of a wind power sequence; meanwhile, a time-varying mode attention mechanism is adopted, the correlation of different time evolution modes of multiple spatial variables is dynamically noticed, and the accurate and effective prediction of the short-term power of the multiple wind power plants is realized.
The technical scheme of the invention is as follows:
a multi-wind-farm short-term power prediction method considering time evolution and space correlation comprises the following steps:
step 1: collecting prediction input data, and collecting historical power data and multi-dimensional meteorological prediction data of a plurality of wind power plants of a target wind power base in a small scale;
step 2: data preprocessing, namely respectively carrying out normalization operation on various data serving as input variables and output variables according to the characteristics of the data;
and step 3: the method comprises the steps of dividing a data set into a training set, a verification set and a test set, firstly, extracting wind power multi-periodicity characteristics by Fourier decomposition based on historical wind power data of the training set, and recording the period length T which is obvious 1 ,T 2 ,...,T u Determining the convolution kernel size of the multi-kernel convolution layer;
and 4, step 4: wind power at d historical moments before the forecasting moments of a plurality of wind power plants and meteorological data forecasting value normalized data at s forecasting moments form a multi-dimensional characteristic input variable, and a depth space-time fusion forecasting model is established by taking wind power values of the s forecasting moments of the plurality of wind power plants as output variables;
and 5: setting model hyper-parameters, initializing weight and bias, setting a loss function, training a deep space-time fusion network to obtain optimal weight and bias parameters, and searching the optimal hyper-parameters of an optimal model by using a grid through a verification set sample;
step 6: and inputting the test sample into a depth space-time fusion model with the optimal hyper-parameters, and performing inverse normalization on the output prediction result to obtain the power prediction result of the multi-wind-field at each moment of the prediction day.
In step 1, the acquired input data consists of l-dimensional known characteristics of n wind power plants at w moments before a prediction moment t, and X belongs to R n×w×l (ii) a For each wind farm there is an input vector x t ＝[P t-d ,P t-d+1 ,...,P t-1 ,Q t ,Q t+1 ,...,Q t+s-1 ]∈R l In which P is t-d ,P t-d+1 ,...,P t-1 Respectively the wind power, Q, of the previous d historical moments t ,Q t+1 ,...,Q t+s-1 And respectively the multi-dimensional meteorological forecast data of the next s forecast moments.
in step 3, based on actual measurement data of small-scale average total wind power of a plurality of wind power plants in a target area of a training set, fourier transform is carried out on the actual measurement data, an amplitude-frequency curve is drawn, the possible difference of periodic laws of wind power in different seasons is considered, multiple seasons and few seasons are divided according to the output characteristic of a wind turbine generator, the periodic laws of the seasons are respectively counted, the proportion of frequency components corresponding to points with larger amplitudes in the amplitude-frequency curve in an original sequence is larger, the frequency points are converted into periods according to sampling frequency, and the multi-period value T of the wind power can be obtained 1 ,T 2 ,…,T u This provides a reference for the subsequent setting of the hyper-parameters of the model.
step 4.1: the depth space-time fusion model firstly sends an input matrix into a Gated Recurrent Unit (GRU), extracts the time sequence information of historical wind power data, and at time t, the GRU receives the current state x t And hidden state h at the previous moment t-1 Output of the network h t Formed by the dynamic control of the update gate and the reset gate. Definition and inputThe associated weight matrix W r ,W u ,W z (ii) a Weight matrix R associated with cyclic concatenation r ,R u ,R z Offset vector b r ,b u ,b z σ is sigmoid activation function, tanh is hyperbolic tangent function, which is a dot product. GRU firstly passes through the hidden state h at the last moment t-1 And input x of the current time t To obtain two gating states, where r t To reset the gate, z t To update the gate:
r t ＝σ(W r x t +R r h t-1 +b r )
z t ＝σ(W z x t +R z h t-1 +b z )
after obtaining the gating signal, the reset gating is used to obtain the hidden state after reset, then the hidden state is spliced with the input and the data range is [ -1,1 ] through the activation function tanh]To obtain
Then selectively forgetting the hidden state transmitted at the previous moment and selectively memorizing the hidden state information containing the current moment to obtain h t The update expression is:
step 4.2: through multi-core convolution layer pair, the hidden state matrix [ h ] obtained after GRU processing t-w+1 ,h t-w+2 ,...,h t-1 ]The row vectors of (a) are processed (w represents the time window length) to extract the multi-periodicity characteristics of the multi-wind farm. Firstly, carrying out convolution operation on row vectors of a hidden state matrix h in k channels by using convolution kernels with u sizes to obtain k different characteristic graphs f map The calculation formula is as follows:
where is the convolution operation, i denotes the ith row vector of the matrix h, concat denotes the feature concatenation operation, K denotes the convolution kernel, with the subscripts denoting the different sizes of the convolution kernel (T) 1 ，T 2 ,…,T u Representing the length of the convolution kernel, 1 representing the width of the convolution kernel) and a channel, and then downsampling (down) the feature map into several sub-blocks through a sliding window to obtain a new feature map f' map (k) In that respect The calculation formula is as follows:
in order to fuse the multi-channel and multi-size convolution operation characteristic diagram, the output vectors of the pooling layer are spliced in the channel direction to obtain
The calculation formula is as follows:
the last step of the multi-core convolutional layer is to linearly map the stitched feature map into a row vector H of a new state matrix H i The calculation formula is as follows:
W f and b f Representing the weights and biases of the mapping process.
Step 4.3: after the time sequence and multi-periodicity time evolution mode information of the wind power sequence is obtained through two steps of processing in steps 4.1 and 4.2, a time-varying mode attention mechanism is used for processing a new hidden state matrix H obtained through GRU and a multi-core convolution layer so as to extract related information of space variables with different time evolution modes, a scoring function f for evaluating the correlation is defined firstly, and the obtained attention weight is normalized, wherein the calculation formula is as follows:
f(H i ,h t )＝(H i ) T W a h t
a i ＝σ(f(H i ,h t ))
wherein W a The weight matrix is obtained through neural network training, and sigma represents a sigmoid activation function; then, the ith row vector (containing time evolution mode information) in the hidden state matrix H and the obtained attention weight a i A weighted sum is performed, which is calculated as follows:
m represents the number of hidden layer neurons.
Step 4.4: finally integrating the characteristic vector v through a dropout layer and a full connection layer t And h t To obtain the final predicted result
The calculation formula is as follows:
h′ t ＝W h h t +W v dropout(v t )
W h ，W v ，W y and representing the corresponding weight matrix, which is also obtained by training the neural network.
Step 6, inputting the test sample into a depth space-time fusion model with the optimal hyper-parameters, and performing inverse normalization on the output prediction result to obtain the power prediction result of each wind power plant at each moment of the prediction day
Wherein n is the number of wind farms and s is a time step predicted later.
Compared with the prior art, the invention has the beneficial effects that:
1) The method considers the time evolution and space related information of multiple wind power plants, simultaneously outputs short-term power prediction results of the multiple wind power plants and multiple time steps, is favorable for realizing grid-connected operation of large-scale wind power data, and has positive influence on the safe and stable operation of a power grid;
2) The method deeply excavates the potential multi-period characteristics of the multi-wind power plant, and combines the extracted multi-period multi-core convolutional layer with the GRU unit for capturing the time sequence dependency relationship, so that the model has the capability of learning the complete time evolution rule of the multi-wind power plant;
3) According to the method, a time-varying mode attention mechanism is introduced, the spatial correlation of multiple wind power plants and multiple time-varying modes is dynamically extracted along with the time, the defect that the spatial dependency of the existing multiple models is extracted statically is overcome, and the prediction stability is enhanced;
4) The practical calculation result shows that the method is reasonable in structure, good in performance on the practical wind power data set, and better in prediction accuracy and robustness compared with the current other hot prediction models.
5) The method not only completely considers the time sequence and periodic time evolution modes of the wind power sequence, but also dynamically takes the spatial correlation information of various time evolution modes of different spatial variables into account, thereby dynamically calculating the spatial correlation of multiple wind power plants; and finally, outputting the power predicted value of each moment of the multi-wind-farm predicted day which simultaneously meets the requirements of time sequence, periodicity and spatial correlation.
Drawings
FIG. 1 is a multi-wind farm short-term power prediction process based on a deep space-time fusion model.
Fig. 2 is a fourier transform amplitude-frequency diagram.
FIG. 3 is a schematic diagram of a gated cycle cell.
FIG. 4 is a diagram illustrating a multi-core convolution operation.
FIG. 5 is a comparison graph of No. 3 wind power plant prediction curve and actual curve in the embodiment of the invention
FIG. 6 is a MAPE error comparison graph (%)
FIG. 7 is an RMSE error comparison graph (MW) of prediction results of different prediction methods of 18 wind power plants in the embodiment of the invention
FIG. 8 is a comparison graph of prediction errors for different cycle characteristics according to an embodiment of the present invention.
Detailed description of the preferred embodiments
The invention provides a multi-wind-farm short-term power prediction framework based on a deep space-time fusion network, which comprises an input module, a time evolution mode tracking module, a space correlation mode attention module and an output module. The input module is used for collecting and preprocessing data, and the objects are historical power data and meteorological prediction data of a multi-wind farm in a target area; the time evolution mode tracking module respectively extracts a time sequence and a multi-periodicity time evolution mode of historical wind power data through the gated circulation unit and the multi-core convolution layer; the spatial correlation mode attention module introduces a time-varying mode attention mechanism to endow correlation weights to different time evolution modes of multiple spatial variables, and can realize transverse comparison of the different time evolution modes of the multiple spatial variables while longitudinally tracking a time evolution rule; and finally, the output module outputs the multi-wind-farm power day-ahead prediction scene which simultaneously meets the requirements of time sequence, periodicity and spatial correlation.
As shown in fig. 1, the method for predicting the short-term power of multiple wind farms based on the deep space-time fusion model includes the following steps:
step 1: acquiring hour-level historical power data and external meteorological information data of a plurality of target wind power plants in a target area, wherein original input data are composed of l-dimensional known characteristics of n wind power plants at w moments before a prediction moment t, and X belongs to R n×w×l . For each wind farm there is an input vector x t ＝[P t-d ,P t-d+1 ,...,P t-1 ,Q t ,Q t+1 ,...,Q t+s-1 ]∈R l In which P is t-d ,P t-d+1 ,...,P t-1 Respectively the wind power, Q, of the previous d historical moments t ,Q t+1 ,...,Q t+s-1 Multi-dimensional meteorological forecast data (such as wind speed, wind direction and air temperature) of the last s forecast moments respectively;
step 2: carrying out normalization processing on the wind power data and the meteorological data acquired in the step 1, wherein the power data are normalized to an interval [0,1 ] by taking the rated capacity of each wind power plant as a reference]Wind speed and temperature adopt a maximum and minimum normalization mode, and wind direction adopts a sin/cos trigonometric function normalization method; the wind power before and after normalization is set as x 1 And
wind speed and temperature of x 2 And
wind direction data is x 3 And
the maximum and minimum values of the wind speed and the temperature sample are x respectively max 、x min Rated capacity of wind power is x N The specific normalization formula is as follows:
and step 3: dividing input data into a training set, a verification set and a test set according to the proportion of 80%, 10% and 10%, performing Fourier transform on actual measurement data of the small-scale average total wind power of a plurality of wind power plants in a target area based on the training set, and drawing an amplitude-frequency curve as shown in FIG. 2. Considering that the periodic laws of wind power in different seasons may have differences, dividing multiple seasons and few seasons according to the output characteristics of the wind turbine generator and respectively counting the periodic laws, wherein the frequency component corresponding to the point with the larger amplitude in the amplitude-frequency curve occupies a larger proportion in the original sequence, and converting the frequency points into the period according to the sampling frequency to obtain the multi-period value T of the wind power 1 ,T 2 ,…,T u This provides a reference for the setting of the model hyper-parameters (multi-kernel convolution kernel size) afterwards;
and 4, step 4: starting to build a deep spatiotemporal fusion model after determining the input and output data, said step 4 comprises the following sub-steps:
step 4.1: the deep space-time fusion model firstly sends an input matrix into a GRU (generalized regression Unit), extracts time sequence information of historical wind power data, and at time t, the GRU receives a current state x t And hidden state h at the previous moment t-1 Output of the network h t Formed by the dynamic control of the update gate and the reset gate. Defining an input-dependent weight matrix W r ,W u ,W z (ii) a Weight matrix R associated with cyclic concatenation r ,R u ,R z Offset vector b r ,b u ,b z σ is sigmoid activation function, tanh is hyperbolic tangent function, which is a dot product. GRU firstly passes through the hidden state h at the last moment t-1 And input x at the current time t To obtain two gating states, where r t To reset the gate, z t To update the door:
r t ＝σ(W r x t +R r h t-1 +b r )
z t ＝σ(W z x t +R z h t-1 +b z )
after obtaining the gating signal, the reset gating is used to obtain the reset hidden state, and then the hidden state is spliced with the input and the data range is [ -1,1 ] by activating the function tanh]To obtain
Then selectively forgetting the hidden state transmitted at the previous moment and selectively memorizing the hidden state information containing the current moment to obtain h t The update expression is:
step 4.2: through the multi-core convolution layer pair, the hidden state matrix [ h ] obtained after GRU processing t-w+1 ,h t-w+2 ,...,h t-1 ]The row vectors of (a) are processed (w represents the time window length) to extract the multi-periodicity characteristics of the multi-wind farm. Firstly, convolution operation is carried out on row vectors of a hidden state matrix h in k channels by using convolution cores with u sizes to obtain k different characteristic diagrams f map The calculation formula is as follows:
where is the convolution operation, i denotes the ith row vector of the matrix h, concat denotes the feature concatenation operation, K denotes the convolution kernel, with the subscripts denoting the different sizes of the convolution kernel (T) 1 ，T 2 ,…,T u Representing the length of the convolution kernel, 1 representing the width of the convolution kernel) and a channel, and then downsampling (down) the feature map into several sub-blocks through a sliding window to obtain a new feature map f' map (k) .1. The The calculation formula is as follows:
in order to fuse the multi-channel and multi-size convolution operation characteristic diagram, the output vectors of the pooling layer are spliced in the channel direction to obtain
The calculation formula is as follows:
the last step of the multi-core convolutional layer is to linearly map the stitched feature map into a row vector H of a new state matrix H i The calculation formula is as follows:
W f and b f Representing the weights and biases of the mapping process.
Step 4.3: after the time evolution mode information of the wind power sequence is obtained through the two steps of processing in the steps 4.1 and 4.2, a time-varying mode attention mechanism is used for processing a new hidden state matrix H obtained through the GRU and the multi-core convolution layer so as to extract related information of space variables with different time evolution modes. Firstly, defining a scoring function f for evaluating the correlation, and carrying out normalization operation on the obtained attention weight, wherein the calculation formula is as follows:
f(H i ,h t )＝(H i ) T W a h t
a i ＝σ(f(H i ,h t ))
wherein W a The weight matrix is obtained by neural network training, and sigma represents a sigmoid activation function; then, the ith row vector (containing time evolution mode information) in the hidden state matrix H and the obtained attention weight a i A weighted sum is performed, which is calculated as follows:
m represents the number of hidden layer neurons.
Step 4.4: finally integrating the characteristic vector v through a dropout layer and a full connection layer t And h t To obtain the final predicted result
The calculation formula is as follows:
h′ t ＝W h h t +W v dropout(v t )
W h ，W v ，W y and representing the corresponding weight matrix, which is also obtained by training the neural network.
And 5: setting hyper-parameters such as the number m of neurons, the length w of a sample time window, the number g of the layers of a time sequence network GRU, the number k of multi-core convolution layer channels and the size of a convolution kernel after the deep space-time fusion model is established; initializing weight and bias, selecting a training sample, taking a mean square error as a loss function, and training the model by adopting an Adam optimization algorithm to obtain optimal weight and bias parameters; inputting the sample of the verification set into a trained deep space-time fusion model, and optimizing the optimal hyper-parameters of the model according to verification errors by adopting grid search, wherein the optimization ranges of some main hyper-parameters are as follows: the number of neurons m {16,32,64,100,128,200,300}; the convolution channel number k is {16,24,32,48}; the number of time sequence network layers g is {1,2,3};
step 6: will measureInputting the sample book into a depth space-time fusion model with optimal hyper-parameters, and performing inverse normalization on the output prediction result to obtain the power prediction result of each wind power plant at each moment of the prediction day
Wherein n is the number of wind farms and s is the time step predicted later.
In this embodiment, the hourly actual measurement power data of 18 wind power plants 2016 of a domestic target wind power base is used as an analysis object, the input of the prediction model is the hourly historical wind power one week ahead of time and the meteorological prediction data (including predicted values of wind speed, wind direction and air temperature per hour) of a prediction day, and the multi-wind-plant power prediction result of the next 24 hours is output. The present embodiment uses the Mean Absolute Percentage Error (MAPE) and the Root Mean Square Error (RMSE) to evaluate the prediction accuracy, and their calculation formula is as follows:
in the formula: n is the number of test samples; y is i And
and respectively predicting the actual value and the predicted value of the wind power of the ith sampling point on the day.
Fig. 5 is a comparison graph of a predicted curve and an actual curve of a wind farm 9 in the embodiment of the present invention, which shows that the method of the present invention performs best in the wind farm and simulates a fluctuation trend of wind power more accurately. And simultaneously has smaller overall error and smaller single-point prediction error. Fig. 6 and 7 are a MAPE error comparison graph and an RMSE error comparison graph of prediction results of different prediction methods for 18 wind farms according to an embodiment of the present invention, respectively, and it can be seen that the method of the present invention has the best prediction accuracy for prediction of the day-ahead wind power. Compared with time sequence neural networks such as TCN, GRU and LSTM, the average improvement rate of MAPE (Imp) of DSTFM is 9.99%, 8.80% and 7.87%, respectively, which shows that only considering the time sequence characteristics of wind power is not enough, and the spatial correlation between a complete time evolution mode and a multi-wind farm needs to be considered. The GRU-CNN is added with a convolution kernel capable of extracting input space characteristics, so that the model has the capability of extracting space-time information, compared with the model, the MAPE average value of the DSTFM is reduced by 5.17%, and the effectiveness of dynamically capturing the space correlation of multiple wind fields and multiple time-varying modes by applying a time-varying mode attention mechanism is reflected. It can also be seen from fig. 6 that MAPE of each model shows difference in different wind farms due to different geographical locations of the multiple wind farms. The MAPE for all models at wind farms No. 4 and 8 is relatively large, indicating that these two windfarms have poor predictability of power. Nevertheless, DSTFM still maintains strong robustness to a plurality of wind farms with different predictability.
FIG. 8 is a comparison graph of the influence of different periodic characteristics of wind power on the prediction result, and three schemes of not considering the periodic characteristics, only considering a single period and considering multi-period combination are compared. The prediction error of the multicycle combination DSTFM model is minimum, and the MAPE error is reduced by 0.75% compared with the period without consideration, which indicates the necessity of taking the periodicity of the wind power data into consideration. Compared with 3 models only considering the monocycle characteristics, the MAPE improvement rate of the model is 0.51%, 0.40% and 0.57% respectively in the cases of 24h, 38h and 80 h. Although the prediction error of the monocycle model is lower than that of a model without introducing the cycle characteristic, the improvement effect is limited. And the stability of the model has a decline potential along with the period extension, which shows that the sensitivity of the model to the time evolution mode change in the period range is poor when the period is longer. And the multi-period combination model can give consideration to the sensitivity to different long and short period characteristics, so that higher prediction accuracy can be obtained.
Claims (7)
1. The method for predicting the short-term power of the multiple wind farms by considering time evolution and space correlation is characterized by comprising the following steps of:
step 1: collecting prediction input data, and collecting historical power data and multi-dimensional meteorological prediction data of a plurality of wind power plants of a target wind power base in a small scale;
step 2: data preprocessing, namely respectively carrying out normalization operation on various data serving as input variables and output variables according to the characteristics of the data;
and 3, step 3: the method comprises the steps of dividing a data set into a training set, a verification set and a test set, firstly, extracting wind power multi-periodicity characteristics based on historical wind power data of the training set by means of Fourier decomposition, and recording the period length T which is obvious 1 ,T 2 ,...,T u Determining the convolution kernel size of the multi-kernel convolution layer;
and 4, step 4: wind power at d historical moments before the forecasting moments of a plurality of wind power plants and meteorological data forecasting value normalized data at s forecasting moments form a multi-dimensional characteristic input variable, and a depth space-time fusion forecasting model is established by taking wind power values of the s forecasting moments of the plurality of wind power plants as output variables;
and 5: setting model hyper-parameters, initializing weight and bias, setting a loss function, training a deep space-time fusion network to obtain optimal weight and bias parameters, and searching the optimal hyper-parameters of an optimal model by using a grid through a verification set sample;
step 6: and inputting the test sample into a depth space-time fusion model with the optimal hyper-parameters, and performing inverse normalization on the output prediction result to obtain the power prediction result of the multi-wind-field at each moment of the prediction day.
2. The method of claim 1, wherein: in step 1, the acquired input data consists of l-dimensional known characteristics of n wind power plants at w moments before a prediction moment t, and X belongs to R n×w×l (ii) a For each wind farm there is an input vector x t ＝[P t-d ,P t-d+1 ,...,P t-1 ,Q t ,Q t+1 ,...,Q t+s-1 ]∈R l In which P is t-d ,P t-d+1 ,...,P t-1 Respectively the wind power, Q, of the previous d historical moments t ,Q t+1 ,...,Q t+s-1 And respectively the multi-dimensional meteorological forecast data of the next s forecast moments.
3. The method of claim 1, wherein: step 2, respectively carrying out normalization processing on the wind power data and the meteorological data acquired in the step 1, wherein the power data are normalized to an interval [0,1 ] by taking the rated capacity of each wind power plant as a reference]Wind speed and temperature adopt a maximum and minimum normalization mode, and wind direction adopts a sin/cos trigonometric function normalization method; the wind power before and after normalization is set as x 1 And
wind speed, temperature x 2 And
wind direction data is x 3 And
the maximum value and the minimum value of the wind speed and the temperature sample are x respectively max 、x min Rated capacity of wind power is x N The specific normalization formula is as follows:
4. the method of claim 1, wherein: in step 3, based on actual measurement data of the small-scale average total wind power of a plurality of wind power plants in a target area of a training set, fourier transform is carried out on the actual measurement data, an amplitude-frequency curve is drawn, the possible difference of periodic laws of wind power in different seasons is considered, multiple seasons and few seasons are divided according to the output characteristics of the wind power generation set, the periodic laws are respectively counted, the occupation ratio of frequency components corresponding to points with larger amplitudes in the amplitude-frequency curve in an original sequence is larger, the frequency points are converted into periods according to sampling frequency, and the multi-period value T of the wind power can be obtained 1 ,T 2 ,…,T u This provides a reference for the subsequent setting of the hyper-parameters of the model.
5. The method of claim 1, wherein: step 4, after input and output data are determined, a depth space-time fusion model is established, and the step 4 comprises the following substeps:
step 4.1: the deep space-time fusion model firstly sends an input matrix into a Gated Recurrent Unit (GRU), extracts time sequence information of historical wind power data, and the GRU receives a current state x at a time t t And hidden state h at the previous moment t-1 Output of the network h t Formed by dynamic control of the update gates and reset gates, defining a weight matrix W associated with the inputs r ,W u ,W z (ii) a Weight matrix R associated with cyclic concatenation r ,R u ,R z Offset vector b r ,b u ,b z σ is sigmoid activation function, tanh is hyperbolic tangent function, Δ is dot product, GRU passes through hidden state h at the previous moment t-1 And input x of the current time t To obtain two gating states, where r t To reset the gate, z t To update the gate:
r t ＝σ(W r x t +R r h t-1 +b r )
z t ＝σ(W z x t +R z h t-1 +b z )
after obtaining the gating signal, the reset gating is used to obtain the reset hidden state, and then the hidden state is spliced with the input and the data range is [ -1,1 ] by activating the function tanh]To obtain
Then selectively forget the hidden state transferred at the previous moment and selectively carry out hidden state information containing the current momentMemorize to obtain h t The update expression is:
step 4.2: through the multi-core convolution layer pair, the hidden state matrix [ h ] obtained after GRU processing t-w+1 ,h t-w+2 ,...,h t-1 ]In order to extract the multi-periodicity characteristics of the multi-wind power plant, firstly convolution operation is carried out on the row vectors of a hidden state matrix h in k channels by using convolution kernels with u sizes to obtain k different characteristic diagrams f map The calculation formula is as follows:
where is the convolution operation, i denotes the ith row vector of the matrix h, concat denotes the feature concatenation operation, K denotes the convolution kernel, with the subscripts denoting the different sizes and channels of the convolution kernel, specifically T 1 ，T 2 ,…,T u Represents the length of the convolution kernel, and 1 represents the width of the convolution kernel; then, a plurality of sub-blocks divided by the feature map are subjected to down-sampling operation through a sliding window, namely down operation in the following formula, so as to obtain a new feature map f m ′ ap (k) The calculation formula is as follows:
in order to fuse multi-channel and multi-size convolution operation characteristic graphs, output vectors of the pooling layer are spliced in the channel direction to obtain
The calculation formula is as follows:
the last step of the multi-core convolutional layer is to linearly map the stitched feature map into a row vector H of a new state matrix H i The calculation formula is as follows:
W f and b f Weights and biases representing the mapping process;
step 4.3: after the time sequence and multi-periodicity time evolution mode information of the wind power sequence is obtained through two steps of processing in steps 4.1 and 4.2, a time-varying mode attention mechanism is used for processing a new hidden state matrix H obtained through GRU and a multi-core convolution layer so as to extract related information of space variables with different time evolution modes, a scoring function f for evaluating the correlation is defined firstly, and the obtained attention weight is normalized, wherein the calculation formula is as follows:
f(H i ,h t )＝(H i ) T W a h t
a i ＝σ(f(H i ,h t ))
wherein W a The weight matrix is obtained through neural network training, and sigma represents a sigmoid activation function; then, the ith row vector containing time evolution mode information in the hidden state matrix H and the obtained attention weight a i A weighted sum is performed, which is calculated as follows:
m represents the number of hidden layer neurons;
step 4.4: finally integrating the characteristic vector v through a dropout layer and a full connection layer t And h t To obtain the final prediction result
The calculation formula is as follows:
W h ，W v ，W y and representing the corresponding weight matrix, which is also obtained by training the neural network.
6. The method of claim 1, wherein: step 5, after the deep space-time fusion model is established in the step 4, setting hyper-parameters such as the number m of neurons, the length w of a sample time window, the number g of layers of a time sequence network GRU, the number k of multi-core convolution layer channels and the size of convolution kernels; initializing weight and bias, selecting a training sample, taking a mean square error as a loss function, and training the model by adopting an Adam optimization algorithm to obtain optimal weight and bias parameters; inputting the sample of the verification set into a trained deep space-time fusion model, and optimizing the optimal hyper-parameters of the model according to verification errors by adopting grid search, wherein the optimization ranges of some main hyper-parameters are as follows: the number of neurons m {16,32,64,100,128,200,300}; the convolution channel number k is {16,24,32,48}; the number of layers g of the time sequence network is 1,2, 3.
7. The method of claim 1, wherein: step 6, inputting the test sample into a depth space-time fusion model with the optimal hyper-parameters, and performing inverse normalization on the output prediction result to obtain the power prediction result of each wind power plant at each moment of the prediction day
Wherein n is the number of wind farms and s is the time step predicted later.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210900738.8A CN115293415A (en) | 2022-07-28 | 2022-07-28 | Multi-wind-farm short-term power prediction method considering time evolution and space correlation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210900738.8A CN115293415A (en) | 2022-07-28 | 2022-07-28 | Multi-wind-farm short-term power prediction method considering time evolution and space correlation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115293415A true CN115293415A (en) | 2022-11-04 |
Family
ID=83823579
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210900738.8A Pending CN115293415A (en) | 2022-07-28 | 2022-07-28 | Multi-wind-farm short-term power prediction method considering time evolution and space correlation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115293415A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115983494A (en) * | 2023-02-10 | 2023-04-18 | 广东工业大学 | Short-term wind power prediction method and system for newly-built small-sample wind power plant |
| CN116050583A (en) * | 2022-12-15 | 2023-05-02 | 生态环境部华南环境科学研究所(生态环境部生态环境应急研究所) | Water environment quality deep learning prediction method coupled with space-time context information |
| CN116417992A (en) * | 2023-03-10 | 2023-07-11 | 华中科技大学 | Fan power prediction model establishment method and application based on space-time attention mechanism |
| CN116451873A (en) * | 2023-06-12 | 2023-07-18 | 中国科学技术大学 | Wind power generation power prediction method and system based on multi-scale double space-time network area |
| CN116703007A (en) * | 2023-08-01 | 2023-09-05 | 中国电力科学研究院有限公司 | Wind power cluster prediction model construction method, power prediction method and device |
| CN116739831A (en) * | 2023-08-16 | 2023-09-12 | 深圳江行联加智能科技有限公司 | Electric energy scheduling method, device and equipment of virtual power plant and storage medium |
| CN116760115A (en) * | 2023-05-30 | 2023-09-15 | 中国南方电网有限责任公司 | Space-time feature fusion power system unit combination optimization method and system thereof |
| CN116826727A (en) * | 2023-06-28 | 2023-09-29 | 河海大学 | Ultra-short-term wind power prediction method and prediction system based on time sequence representation and multistage attention |
| CN116894384A (en) * | 2023-07-11 | 2023-10-17 | 湖北工业大学 | Multi-fan wind speed space-time prediction method and system |
| CN116937579A (en) * | 2023-09-19 | 2023-10-24 | 太原理工大学 | Wind power interval prediction considering space-time correlation and interpretable method thereof |
| CN117114192A (en) * | 2023-08-29 | 2023-11-24 | 淮阴工学院 | Offshore wind power prediction method and device based on multi-scale space-time diagram transformation network |
| CN117220278A (en) * | 2023-09-13 | 2023-12-12 | 四川大学 | Long-term wind power prediction method considering non-stationary space-time dependence |
| CN117595238A (en) * | 2023-10-31 | 2024-02-23 | 北京爱特瑞科软件开发有限责任公司 | Training method of wind power prediction model and wind power prediction method |
| CN117878928A (en) * | 2024-03-12 | 2024-04-12 | 国能日新科技股份有限公司 | Wind power prediction method and device based on deep learning |
| CN117893362A (en) * | 2024-03-18 | 2024-04-16 | 广东工业大学 | Multi-time-space-scale offshore wind power characteristic screening and enhanced power prediction method |
| CN117937607A (en) * | 2024-03-21 | 2024-04-26 | 太原理工大学 | Power system safety and stability analysis method considering space-time law of wind turbine generator |
| CN118195361A (en) * | 2024-05-17 | 2024-06-14 | 国网吉林省电力有限公司经济技术研究院 | Big data-based energy management method and system |
| CN118611059A (en) * | 2024-08-07 | 2024-09-06 | 浙江大学海南研究院 | Offshore wind power prediction method of attention mechanism double-channel neural network |
-
2022
- 2022-07-28 CN CN202210900738.8A patent/CN115293415A/en active Pending
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116050583B (en) * | 2022-12-15 | 2024-05-28 | 生态环境部华南环境科学研究所(生态环境部生态环境应急研究所) | Water environment quality deep learning prediction method coupled with space-time context information |
| CN116050583A (en) * | 2022-12-15 | 2023-05-02 | 生态环境部华南环境科学研究所(生态环境部生态环境应急研究所) | Water environment quality deep learning prediction method coupled with space-time context information |
| CN115983494A (en) * | 2023-02-10 | 2023-04-18 | 广东工业大学 | Short-term wind power prediction method and system for newly-built small-sample wind power plant |
| CN115983494B (en) * | 2023-02-10 | 2023-09-12 | 广东工业大学 | Short-term wind power prediction method and system for newly-built small-sample wind power plant |
| CN116417992A (en) * | 2023-03-10 | 2023-07-11 | 华中科技大学 | Fan power prediction model establishment method and application based on space-time attention mechanism |
| CN116417992B (en) * | 2023-03-10 | 2024-03-19 | 华中科技大学 | Fan power prediction model establishment method and application based on space-time attention mechanism |
| CN116760115B (en) * | 2023-05-30 | 2024-04-12 | 中国南方电网有限责任公司 | Space-time feature fusion power system unit combination optimization method and system thereof |
| CN116760115A (en) * | 2023-05-30 | 2023-09-15 | 中国南方电网有限责任公司 | Space-time feature fusion power system unit combination optimization method and system thereof |
| CN116451873A (en) * | 2023-06-12 | 2023-07-18 | 中国科学技术大学 | Wind power generation power prediction method and system based on multi-scale double space-time network area |
| CN116451873B (en) * | 2023-06-12 | 2023-10-20 | 中国科学技术大学 | Wind power generation power prediction method and system based on multi-scale double space-time network area |
| CN116826727A (en) * | 2023-06-28 | 2023-09-29 | 河海大学 | Ultra-short-term wind power prediction method and prediction system based on time sequence representation and multistage attention |
| CN116826727B (en) * | 2023-06-28 | 2024-02-23 | 河海大学 | Ultra-short-term wind power prediction method and prediction system based on time sequence representation and multistage attention |
| CN116894384A (en) * | 2023-07-11 | 2023-10-17 | 湖北工业大学 | Multi-fan wind speed space-time prediction method and system |
| CN116894384B (en) * | 2023-07-11 | 2024-05-07 | 湖北工业大学 | Multi-fan wind speed space-time prediction method and system |
| CN116703007A (en) * | 2023-08-01 | 2023-09-05 | 中国电力科学研究院有限公司 | Wind power cluster prediction model construction method, power prediction method and device |
| CN116703007B (en) * | 2023-08-01 | 2023-12-15 | 中国电力科学研究院有限公司 | Wind power cluster prediction model construction method, power prediction method and device |
| CN116739831A (en) * | 2023-08-16 | 2023-09-12 | 深圳江行联加智能科技有限公司 | Electric energy scheduling method, device and equipment of virtual power plant and storage medium |
| CN117114192A (en) * | 2023-08-29 | 2023-11-24 | 淮阴工学院 | Offshore wind power prediction method and device based on multi-scale space-time diagram transformation network |
| CN117220278A (en) * | 2023-09-13 | 2023-12-12 | 四川大学 | Long-term wind power prediction method considering non-stationary space-time dependence |
| CN117220278B (en) * | 2023-09-13 | 2024-03-19 | 四川大学 | Long-term wind power prediction method considering non-stationary space-time dependence |
| CN116937579B (en) * | 2023-09-19 | 2023-12-01 | 太原理工大学 | Wind power interval prediction considering space-time correlation and interpretable method thereof |
| CN116937579A (en) * | 2023-09-19 | 2023-10-24 | 太原理工大学 | Wind power interval prediction considering space-time correlation and interpretable method thereof |
| CN117595238A (en) * | 2023-10-31 | 2024-02-23 | 北京爱特瑞科软件开发有限责任公司 | Training method of wind power prediction model and wind power prediction method |
| CN117878928A (en) * | 2024-03-12 | 2024-04-12 | 国能日新科技股份有限公司 | Wind power prediction method and device based on deep learning |
| CN117878928B (en) * | 2024-03-12 | 2024-05-14 | 国能日新科技股份有限公司 | Wind power prediction method and device based on deep learning |
| CN117893362A (en) * | 2024-03-18 | 2024-04-16 | 广东工业大学 | Multi-time-space-scale offshore wind power characteristic screening and enhanced power prediction method |
| CN117937607A (en) * | 2024-03-21 | 2024-04-26 | 太原理工大学 | Power system safety and stability analysis method considering space-time law of wind turbine generator |
| CN117937607B (en) * | 2024-03-21 | 2024-05-31 | 太原理工大学 | Power system safety and stability analysis method considering space-time law of wind turbine generator |
| CN118195361A (en) * | 2024-05-17 | 2024-06-14 | 国网吉林省电力有限公司经济技术研究院 | Big data-based energy management method and system |
| CN118611059A (en) * | 2024-08-07 | 2024-09-06 | 浙江大学海南研究院 | Offshore wind power prediction method of attention mechanism double-channel neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115293415A (en) | Multi-wind-farm short-term power prediction method considering time evolution and space correlation | |
| CN111667098B (en) | Wind power station output power prediction method based on multi-model combination optimization | |
| CN107194495B (en) | Photovoltaic power longitudinal prediction method based on historical data mining | |
| CN108022001A (en) | Short term probability density Forecasting Methodology based on PCA and quantile estimate forest | |
| CN112561058B (en) | Short-term photovoltaic power prediction method based on Stacking-integrated learning | |
| CN113052469B (en) | Method for calculating wind-solar-water-load complementary characteristic of small hydropower area lacking measurement runoff | |
| CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
| CN112100911A (en) | Solar radiation prediction method based on deep BISLTM | |
| CN116187540B (en) | Wind power station ultra-short-term power prediction method based on space-time deviation correction | |
| CN115099500A (en) | Water level prediction method based on weight correction and DRSN-LSTM model | |
| He et al. | A power forecasting approach for PV plant based on irradiance index and LSTM | |
| CN113962441A (en) | Short-term irradiance prediction method and prediction system based on historical data analysis | |
| CN116894504A (en) | Wind power cluster power ultra-short-term prediction model establishment method | |
| CN115481788A (en) | Load prediction method and system for phase change energy storage system | |
| Wibawa et al. | Long Short-Term Memory to Predict Unique Visitors of an Electronic Journal | |
| Shi et al. | Deep-learning-based wind speed forecasting considering spatial–temporal correlations with adjacent wind turbines | |
| CN118134284A (en) | Deep learning wind power prediction method based on multi-stage attention mechanism | |
| Hou et al. | A hybrid machine learning forecasting model for photovoltaic power | |
| Zhou et al. | Short-term wind power prediction based on EMD-LSTM | |
| Kermia et al. | Photovoltaic power prediction using a recurrent neural network RNN | |
| Li et al. | The application of deep neural network algorithm considering multi-dimensional meteorological feature extraction in short-term wind speed prediction | |
| Ding et al. | Photovoltaic array power prediction model based on EEMD and PSO-KELM | |
| CN115296298A (en) | Wind power plant power prediction method | |
| Dhakal et al. | Deep Learning Model with Probability Density Function and Feature Engineering for Short Term Wind Speed Prediction | |
| Gao et al. | Multiple wind speed forecast using simple recurrent unit |
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 |