CN117810997A - Short-term wind power prediction method based on space-time correlation - Google Patents
Short-term wind power prediction method based on space-time correlation Download PDFInfo
- Publication number
- CN117810997A CN117810997A CN202410235440.9A CN202410235440A CN117810997A CN 117810997 A CN117810997 A CN 117810997A CN 202410235440 A CN202410235440 A CN 202410235440A CN 117810997 A CN117810997 A CN 117810997A
- Authority
- CN
- China
- Prior art keywords
- data
- wind power
- encoder
- decoder
- layer
- 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 51
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 10
- 230000001932 seasonal effect Effects 0.000 claims description 25
- 230000002159 abnormal effect Effects 0.000 claims description 24
- 238000004364 calculation method Methods 0.000 claims description 17
- 238000000354 decomposition reaction Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 12
- 239000006185 dispersion Substances 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a short-term wind power prediction method based on space-time correlation, which comprises the following steps: acquiring historical power of a wind power plant in a certain area and historical data of wind power plant influence factors to be tested, and establishing a characteristic database; the data is subjected to dimension reduction while the data similarity is reserved by using a t-distribution random neighbor embedding algorithm; extracting local features of the processed wind power data by using CNN; and using a FEDformer model with a main framework of an encoder-decoder, using an MOE module in the encoder to decompose an input signal, weighting a FEB module, using a FEA module in the decoder to execute cross attention operation on the signal, extracting correlation and similarity in the signal, and realizing self-adaptive learning and prediction of the model to obtain a final prediction result. The invention can provide a foundation for balancing the power supply and load demand of the system, prevent the system from malfunctioning due to wind power access to the power grid, and improve the stability of the power system.
Description
Technical Field
The invention relates to the field of wind power prediction of new energy power generation and network access, in particular to a short-term wind power prediction method based on space-time correlation.
Background
With the continuous development of new energy sources such as solar energy, wind power and the like, the ratio of the new energy sources in the power system is continuously improved, and the novel energy source plays an important role in relieving the energy problem in China. Meanwhile, under the 'double carbon' target, new energy sources such as solar energy, wind energy and the like are rapidly developed, and high-proportion new energy source access to a power grid will become a trend. The wind power in China has the characteristics of abundant reserves, wide distribution, strong availability and the like, the wind power is rapidly developed and applied in China, but the wind power output has volatility and randomness, so that the electric energy on a plurality of time scales of a power grid is unbalanced, and great challenges are brought to power grid dispatching. Therefore, the rapid and accurate wind power prediction is significant for the supply and demand balance and the power supply safety of the power system.
Disclosure of Invention
The invention aims to: the invention aims to provide a short-term wind power prediction method based on space-time correlation, so as to provide a basis for balancing power supply and load demand of a system, prevent the system from being failed due to wind power access to a power grid, and ensure safe and stable operation of the power system.
The technical scheme is as follows: the invention discloses a short-term wind power prediction method based on space-time correlation, which comprises the following steps:
(1) Acquiring historical power of a wind power plant in a certain area and historical data of wind power plant influence factors to be tested, establishing a characteristic database, identifying and removing abnormal values in original data, filling missing values in the original data, and normalizing the data by using a z-score; the wind farm influence factors to be measured comprise wind speed, wind direction, temperature and humidity. The method comprises the following steps:
and (1.1) sorting historical power of a plurality of wind power plants in a certain area and historical data of meteorological factors of a power station to be tested, identifying abnormal values in a sample set, removing the abnormal values, and then complementing missing values in original data.
(1.1.1) applying a Laida criterion method to give a confidence probability, determining a confidence limit according to the confidence probability, judging whether the data is an abnormal value by taking the confidence limit as a standard, and if the data error exceeds the limit, regarding the data as the abnormal value and rejecting the abnormal value, thereby finally obtaining the wind power plant historical power data and the influence factor historical data.
(1.1.2) on the basis of eliminating abnormal values, carrying out filling by adopting an average value method aiming at the situation of non-multiple continuous data missing, and processing by adopting a direct deleting method aiming at the situation of multiple continuous data missing; the average value method has the formula as follows:
;
wherein x is i For the ith missing data, x i-1 And x i+1 For data before and after the missing value, if x i+1 Is a missing value, then take x backward i+2 。
(1.1.3) normalizing the washed historical Power data using the z-score, and recording the sample sequence as x 1 ,x 2 ,…,x n The normalization formula is as follows:
;
wherein x is i Is data before normalization, y i Is the corresponding standardized data of the data set,is the average of the series, and a new sequence y is generated 1 ,y 2 ,…,y n Is data with a mean value of 0, a variance of 1 and no dimension.
(2) And (3) using a t-distribution random neighbor embedding algorithm to reduce the dimension of the data while retaining the similarity of the data. The method comprises the following steps:
performing feature extraction on historical operation data of the wind power plant by using a t-distribution random neighbor embedding algorithm;
note s= { x 1 ,x 2 ,…,x n Data set running for n high-dimensional systems, high-dimensional data points obey Gaussian distribution between every two, and P is used j|i Describing the conditional probability of Gaussian center similarity, the calculation formula is as follows:
;
wherein,data point x i Gaussian distribution variance of (c);
according to this, a joint probability distribution P of any two data points in space is calculated ij The calculation formula is as follows:
;
wherein P is i|j Is at a given data point x i Conditions of (2)In the case of probability, x j Probability as a neighboring point; p (P) j|i Is at a given data point x j In the case of conditional probability of (2), x will be i Probability as a neighboring point;
in the low-dimensional space, the notation t= { y 1 ,y 2 ,…,y n Mapping of the set S to t distribution subject to degree of freedom l, Q ij The calculation formula of the joint probability distribution of any two feature matrixes of the system operation data is as follows:
the method comprises the steps of carrying out a first treatment on the surface of the Then, using Kullback-Leibler dispersion D KL Representing the correctness of the mapping, the calculation formula is:
;
wherein P is ij Is the joint probability distribution of any two data points in space, Q ij The probability distribution is combined by any two feature matrixes of the system operation data;
and finally, iteratively solving the minimum dispersion by using a gradient descent method, and improving the accuracy of the meteorological feature matrix.
(3) Extracting local features of the processed wind power data by using CNN to generate more important information; the method comprises the following steps:
(3.1) the input of the CNN is multi-feature data, and effective nonlinear local features of the input time sequence are extracted through a convolution layer; dividing data into k days in a unit of a day, n pieces of data each day, wherein each piece of data comprises m meteorological factors, an n multiplied by m multiplied by k matrix is formed as input of a CNN model, and the output of a CNN convolution layer is recorded asThe convolution layer output formula is as follows:
wherein,to activate the function, k is the sliding window size, w n,m Weights for n rows and m columns of convolution kernel, X i+n,j+m B, inputting values of nth row and mth column of wind power data feature matrix m,n Is a convolution kernel deviation.
And (3.2) in the CNN pooling layer, performing data sampling by using a filter and a sliding window, reducing the characteristic size of the data, reducing network parameters, and finally inputting the data into the FEDformer model through the full connection layer.
(4) The method comprises the steps of using an FEDformer model with a main framework of an encoder-decoder, decomposing an input signal through an MOE module of the encoder, weighting frequency domain features by an FEB module to enable the high frequency features to obtain higher weights, performing cross attention operation on signals from the encoder and the decoder by FEA in the decoder, extracting correlation and similarity in the signals, realizing self-adaptive learning and prediction of the model, and obtaining a final prediction result. The method comprises the following steps:
(4.1) the encoder adopts a multi-layered structure,the MOE Decomp layer in the encoder is responsible for decomposing the input signal into a seasonal component and a trend component, discarding the trend component, retaining the seasonal component, and passing it to the next layer of learning; the FEB module is responsible for extracting the frequency domain information of the signal, integrating the frequency domain information into a model, weighting the frequency characteristics to ensure that the high frequency characteristics can be weighted more, and finally feeding the processed seasonal components to a decoder.
The main formulas of the encoder are as follows:
;
;
;
wherein,representing the output of the layer I encoder,/I>Is a sequence of histories that are embedded,representing the seasonal component after the ith decomposition block in the first layer.
(4.2) the decoder also adopts a multi-layer structure,the input in the decoder will also go through three MOE Decomp decomposition layers, which decompose the signal into trend components and seasonal components, and pass the seasonal components to the next layer for learning step by step as the encoder; the FEB is responsible for extracting effective frequency domain information in the time sequence, has the same function as a self-care block, and weights high-frequency characteristics so as to better capture detailed information of the time sequence; the FEA is responsible for exchanging information between the encoder and the decoder, learning the internal relation between the two modules through cross attention operation, extracting the correlation and similarity between signals, and realizing the self-adaptive learning and prediction of the model;
the main formulas of the decoder are as follows:
;
;
;
;
wherein,representing the output of the layer I encoder,/I>Respectively representing the seasonal component and the trend component after the i-th component of the first layer is separated, the ++>Respectively represent the trend of the ith extraction +.>Is a projection of (2);
the final prediction result is the sum of two refinement decomposition components, and the formula is:
;
wherein,is to transform the depth-transformed seasonal component +.>Projected to the target.
A computer storage medium having stored thereon a computer program which when executed by a processor implements a short-term wind power prediction method based on spatio-temporal correlation as described above.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing a short-term wind power prediction method based on spatio-temporal correlation as described above when executing the computer program.
The beneficial effects are that: compared with the prior art, the invention has the following advantages:
1. according to the short-term wind power generation prediction model based on space-time correlation, the effects of time sequence features and external influence factors are comprehensively considered, and feature selection is carried out through a t-SNE algorithm.
2. The invention uses the FEDformer model, combines the transform and the seasonal trend decomposition method, and adopts a brand new architecture of the encoder-decoder, and applies the transform to the frequency domain by introducing Fourier analysis, thereby not only reducing the calculation cost of the transform, but also improving the prediction precision.
Drawings
FIG. 1 is a flow chart of the steps of the method of the present invention.
Fig. 2 is a schematic diagram of an FEDformer model.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
As shown in FIG. 1, the short-term wind power prediction method based on space-time correlation comprises the following steps:
(1) Acquiring historical power of a wind power plant in a certain area and historical data of wind power plant influence factors to be tested, establishing a characteristic database, identifying and removing abnormal values in original data, filling missing values in the original data, and normalizing the data by using a z-score;
and (1.1) sorting historical power of a plurality of wind power plants in a certain area and historical data of meteorological factors of a power station to be tested, identifying abnormal values in a sample set, removing the abnormal values, and then complementing missing values in original data.
(1.1.1) applying a Laida criterion method to give a confidence probability, determining a confidence limit according to the confidence probability, judging whether the data is an abnormal value by taking the confidence limit as a standard, and if the data error exceeds the limit, regarding the data as the abnormal value and rejecting the abnormal value, thereby finally obtaining the wind power plant historical power data and the influence factor historical data.
(1.1.2) on the basis of eliminating abnormal values, carrying out filling by adopting an average value method aiming at the situation of non-multiple continuous data missing, and processing by adopting a direct deleting method aiming at the situation of multiple continuous data missing; the average value method has the formula as follows:
wherein x is i For the ith missing data, x i-1 And x i+1 For data before and after the missing value, if x i+1 Is a missing value, then take x backward i+2 。
(1.1.3) normalizing the washed historical Power data using the z-score, and recording the sample sequence as x 1 ,x 2 ,…,x n The normalization formula is as follows:
;
wherein x is i Is data before normalization, y i Is the corresponding standardized data of the data set,is the average of the series, and a new sequence y is generated 1 ,y 2 ,…,y n Is data with a mean value of 0, a variance of 1 and no dimension.
(2) And (3) using a t-distribution random neighbor embedding algorithm to reduce the dimension of the data while retaining the similarity of the data. The method comprises the following steps:
performing feature extraction on historical operation data of the wind power plant by using a t-distribution random neighbor embedding algorithm;
note s= { x 1 ,x 2 ,…,x n Data set running for n high-dimensional systems, high-dimensional data points obey Gaussian distribution between every two, and P is used j|i Describing the conditional probability of Gaussian center similarity, the calculation formula is as follows:
;
wherein,data point x i Gaussian distribution variance of (c);
according to this, a joint probability distribution P of any two data points in space is calculated ij The calculation formula is as follows:
;
wherein P is i|j Is at a given data point x i In the case of conditional probability of (2), x will be j Probability as a neighboring point; p (P) j|i Is at a given data point x j In the case of conditional probability of (2), x will be i Probability as a neighboring point;
in the low-dimensional space, the notation t= { y 1 ,y 2 ,…,y n Mapping of the set S to t distribution subject to degree of freedom l, Q ij The calculation formula of the joint probability distribution of any two feature matrixes of the system operation data is as follows:
;
then, using Kullback-Leibler dispersion D KL Representing the correctness of the mapping, the calculation formula is:
;
wherein P is ij Is the joint probability distribution of any two data points in space, Q ij The probability distribution is combined by any two feature matrixes of the system operation data;
and finally, iteratively solving the minimum dispersion by using a gradient descent method, and improving the accuracy of the meteorological feature matrix.
(3) Extracting local features of the processed wind power data by using CNN to generate more important information;
(3.1) the input of the CNN is multi-feature data, and effective nonlinear local features of the input time sequence are extracted through a convolution layer; dividing data into k days in a unit of a day, n pieces of data each day, wherein each piece of data comprises m meteorological factors, an n multiplied by m multiplied by k matrix is formed as input of a CNN model, and the output of a CNN convolution layer is recorded asThe convolution layer output formula is as follows:
wherein,to activate the function, k is the sliding window size, w n,m Weights for n rows and m columns of convolution kernel, X i+n,j+m B, inputting values of nth row and mth column of wind power data feature matrix m,n Is a convolution kernel deviation.
And (3.2) in the CNN pooling layer, performing data sampling by using a filter and a sliding window, reducing the characteristic size of the data, reducing network parameters, and finally inputting the data into the FEDformer model through the full connection layer.
(4) The method comprises the steps of using an FEDformer model with a main framework of an encoder-decoder, decomposing an input signal through an MOE module of the encoder, weighting frequency domain features by an FEB module to enable the high frequency features to obtain higher weights, performing cross attention operation on signals from the encoder and the decoder by FEA in the decoder, extracting correlation and similarity in the signals, realizing self-adaptive learning and prediction of the model, and obtaining a final prediction result.
Fig. 2 is a schematic diagram of an FEDformer model according to the present invention, which mainly includes a workflow of the FEDformer model, and the details are as follows:
(4.1) performing outlier rejection by using the Laida criterion method. The specific operation is as follows:
let the measured be measured with equal precision and independently obtain x 1 ,x 2 ,x 3 ,……,x n Calculate the arithmetic mean valueResidual error->And calculate the standard deviation +.>If a certain measured value->Residual error of (2)Satisfies the following formula
Then consider asShould be rejected.
And then carrying out missing data complementation by using an average value method, wherein the specific operation is as follows:
on the basis of eliminating abnormal values, the method is used for supplementing and correcting the abnormal values by adopting an average value method aiming at the condition of not deleting a plurality of continuous data, and the method is used for processing the abnormal values by adopting a direct deleting method aiming at the condition of deleting a plurality of continuous data; the average value method has the formula as follows:
wherein x is i For the ith missing data, x i-1 And x i+1 For data before and after the missing value, if x i+1 Is a missing value, then take x backward i+2 。
And then carrying out data standardization processing:
;
wherein x is i Is data before normalization, y i Is the corresponding standardized data of the data set,is the average of the series, and a new sequence y is generated 1 ,y 2 ,…,y n Is data with a mean value of 0, a variance of 1 and no dimension.
And (4.2) using a t-distribution random neighbor embedding algorithm to reduce the dimension of the data while retaining the similarity of the data. The method comprises the following steps:
performing feature extraction on historical operation data of the wind power plant by using a t-distribution random neighbor embedding algorithm;
note s= { x 1 ,x 2 ,…,x n Data set running for n high-dimensional systems, high-dimensional data points obey Gaussian distribution between every two, and P is used j|i Describing the conditional probability of Gaussian center similarity, the calculation formula is as follows:
;
wherein,data point x i Gaussian distribution variance of (c);
according to this, a joint probability distribution P of any two data points in space is calculated ij The calculation formula is as follows:
;
wherein P is i|j Is at a given data point x i In the case of conditional probability of (2), x will be j Probability as a neighboring point; p (P) j|i Is at a given data point x j In the case of conditional probability of (2), x will be i Probability as a neighboring point;
in the low-dimensional space, the notation t= { y 1 ,y 2 ,…,y n Mapping of the set S to t distribution subject to degree of freedom l, Q ij The calculation formula of the joint probability distribution of any two feature matrixes of the system operation data is as follows:
;
then, using Kullback-Leibler dispersion D KL Representing that the mapping is correctThe calculation formula is as follows:
;
wherein P is ij Is the joint probability distribution of any two data points in space, Q ij The probability distribution is combined by any two feature matrixes of the system operation data;
and finally, iteratively solving the minimum dispersion by using a gradient descent method, and improving the accuracy of the meteorological feature matrix.
(4.3) extracting local characteristics of the processed wind power data by using CNN to generate more important information:
the CNN is input as multi-feature data, and effective nonlinear local features of an input time sequence are extracted through a convolution layer; dividing data into k days in a unit of a day, n pieces of data each day, wherein each piece of data comprises m meteorological factors, an n multiplied by m multiplied by k matrix is formed as input of a CNN model, and the output of a CNN convolution layer is recorded asThe convolution layer output formula is as follows:
wherein,to activate the function, k is the sliding window size, w n,m Weights for n rows and m columns of convolution kernel, X i+n,j+m B, inputting values of nth row and mth column of wind power data feature matrix m,n Is a convolution kernel deviation.
In the CNN pooling layer, a filter and a sliding window are utilized to sample data, the characteristic size of the data is reduced, network parameters are reduced, and finally the data is input into the FEDformer model through the full connection layer.
(4.4) first combining a transform with a seasonal trend decomposition method to build a FEDformer prediction model mainly configured as an encoder-decoder, the model comprising a Frequency Enhancement Block (FEB), a Frequency Enhancement Attention (FEA), a periodic trend decomposition (PD) and a Forward Propagation (FP):
the encoder adopts a multi-layer structure,the MOE Decomp layer in the encoder is responsible for decomposing the input signal into a seasonal component and a trend component, discarding the trend component, retaining the seasonal component, and passing it to the next layer of learning; the FEB module is responsible for extracting the frequency domain information of the signal, integrating the frequency domain information into a model, weighting the frequency characteristics to ensure that the high frequency characteristics can be weighted more, and finally feeding the processed seasonal components to a decoder.
The main formulas of the encoder are as follows:
;
;
;
wherein,representing the output of the layer I encoder,/I>Is a sequence of histories that are embedded,representing the seasonal component after the ith decomposition block in the first layer.
The decoder also adopts a multi-layer structure,the input in the decoder will also go through three MOE Decomp decomposition layers which decompose the signal into trend components and seasonal components, which will be the same as the encoderTransferring to the next layer for learning step by step; the FEB is responsible for extracting effective frequency domain information in the time sequence, has the same function as a self-care block, and weights high-frequency characteristics so as to better capture detailed information of the time sequence; FEA is responsible for exchanging information between the encoder and the decoder, learning the internal relation between the two modules through cross attention operation, extracting the correlation and similarity between signals, and realizing the self-adaptive learning and prediction of the model.
The main formulas of the decoder are as follows:
;
;
;
;
wherein,representing the output of the layer I encoder,/I>Respectively representing the seasonal component and the trend component after the i-th component of the first layer is separated, the ++>Respectively represent the trend of the ith extraction +.>Is a projection of (2);
the final prediction result is the sum of two refinement decomposition components, and the formula is:
;
wherein,is to transform the depth-transformed seasonal component +.>Projected to the target.
Claims (8)
1. A short-term wind power prediction method based on space-time correlation is characterized by comprising the following steps:
(1) Acquiring historical power of a wind power plant in a certain area and historical data of wind power plant influence factors to be tested, establishing a characteristic database, identifying and removing abnormal values in original data, filling missing values in the original data, and normalizing the data by using a z-score;
(2) Using a t-distribution random neighbor embedding algorithm, and reducing the dimension of the data while keeping the similarity of the data;
(3) Extracting local features of the processed wind power data by using CNN to generate more important information;
(4) The method comprises the steps of using an FEDformer model with a main framework of an encoder-decoder, decomposing an input signal through an MOE module of the encoder, weighting frequency domain features by an FEB module to enable the high frequency features to obtain higher weights, performing cross attention operation on signals from the encoder and the decoder by FEA in the decoder, extracting correlation and similarity in the signals, realizing self-adaptive learning and prediction of the model, and obtaining a final prediction result.
2. The method for predicting short-term wind power based on space-time correlation according to claim 1, wherein the wind farm influence factors to be measured in step (1) include wind speed, wind direction, temperature and humidity.
3. The short-term wind power prediction method based on space-time correlation according to claim 1, wherein the step (1) specifically comprises:
the method comprises the steps of (1.1) sorting historical power of a plurality of wind power plants in a certain area and historical data of meteorological factors of a power station to be tested, identifying abnormal values in a sample set, removing the abnormal values, and then complementing missing values in original data;
(1.1.1) applying a Laida criterion method to give a confidence probability, determining a confidence limit according to the confidence probability, judging whether the data is an abnormal value or not by taking the confidence limit as a standard, and if the data error exceeds the limit, regarding the data as the abnormal value and removing the abnormal value, so as to finally obtain wind power plant historical power data and influence factor historical data;
(1.1.2) on the basis of eliminating abnormal values, carrying out filling by adopting an average value method aiming at the situation of non-multiple continuous data missing, and processing by adopting a direct deleting method aiming at the situation of multiple continuous data missing; the average value method has the formula as follows:
;
wherein x is i For the ith missing data, x i-1 And x i+1 For data before and after the missing value, if x i+1 Is a missing value, then take x backward i+2 ;
(1.1.3) normalizing the washed historical Power data using the z-score, and recording the sample sequence as x 1 ,x 2 ,…,x n The normalization formula is as follows:
;
wherein x is i Is data before normalization, y i Is the corresponding standardized data of the data set,is the average of the series, and a new sequence y is generated 1 ,y 2 ,…,y n Is data with a mean value of 0, a variance of 1 and no dimension.
4. The short-term wind power prediction method based on space-time correlation according to claim 1, wherein the step (2) specifically comprises:
performing feature extraction on historical operation data of the wind power plant by using a t-distribution random neighbor embedding algorithm;
note s= { x 1 ,x 2 ,…,x n Data set running for n high-dimensional systems, high-dimensional data points obey Gaussian distribution between every two, and P is used j|i Describing the conditional probability of Gaussian center similarity, the calculation formula is as follows:
;
wherein,data point x i Gaussian distribution variance of (c);
according to this, a joint probability distribution P of any two data points in space is calculated ij The calculation formula is as follows:
;
wherein P is i|j Is at a given data point x i In the case of conditional probability of (2), x will be j Probability as a neighboring point; p (P) j|i Is at a given data point x j In the case of conditional probability of (2), x will be i Probability as a neighboring point;
in the low-dimensional space, the notation t= { y 1 ,y 2 ,…,y n Mapping of the set S to t distribution subject to degree of freedom l, Q ij The calculation formula of the joint probability distribution of any two feature matrixes of the system operation data is as follows:
the method comprises the steps of carrying out a first treatment on the surface of the Then, using Kullback-Leibler dispersion D KL Representing mapping correctnessThe calculation formula is as follows:
;
wherein P is ij Is the joint probability distribution of any two data points in space, Q ij The probability distribution is combined by any two feature matrixes of the system operation data;
and finally, iteratively solving the minimum dispersion by using a gradient descent method, and improving the accuracy of the meteorological feature matrix.
5. The short-term wind power prediction method based on space-time correlation according to claim 1, wherein the step (3) specifically comprises:
(3.1) the input of the CNN is multi-feature data, and effective nonlinear local features of the input time sequence are extracted through a convolution layer; dividing data into k days in a unit of a day, n pieces of data each day, wherein each piece of data comprises m meteorological factors, an n multiplied by m multiplied by k matrix is formed as input of a CNN model, and the output of a CNN convolution layer is recorded asThe convolution layer output formula is as follows:
;
wherein,to activate the function, k is the sliding window size, w n,m Weights for n rows and m columns of convolution kernel, X i+n,j+m B, inputting values of nth row and mth column of wind power data feature matrix m,n Is a convolution kernel deviation;
and (3.2) in the CNN pooling layer, performing data sampling by using a filter and a sliding window, reducing the characteristic size of the data, reducing network parameters, and finally inputting the data into the FEDformer model through the full connection layer.
6. The short-term wind power prediction method based on space-time correlation according to claim 1, wherein the step (4) specifically comprises:
(4.1) the encoder adopts a multi-layered structure,the MOE Decomp layer in the encoder is responsible for decomposing the input signal into a seasonal component and a trend component, discarding the trend component, retaining the seasonal component, and passing it to the next layer of learning; the FEB module is responsible for extracting frequency domain information of signals, integrating the frequency domain information into a model, weighting frequency characteristics to ensure that the high-frequency characteristics can obtain larger weight, and finally feeding processed seasonal components to a decoder;
the main formulas of the encoder are as follows:
;
;
;
wherein,representing the output of the layer I encoder,/I>Is an embedded history sequence,/>Representing seasonal components after the ith decomposition block in the first layer;
(4.2) the decoder also adopts a multi-layer structure,the input in the decoder will also go through three MOE Decomp decomposition layers, which decompose the signal into trend components and seasonal components, and pass the seasonal components to the next layer for learning step by step as the encoder; the FEB is responsible for extracting effective frequency domain information in the time sequence, has the same function as a self-care block, and weights high-frequency characteristics so as to better capture detailed information of the time sequence; the FEA is responsible for exchanging information between the encoder and the decoder, learning the internal relation between the two modules through cross attention operation, extracting the correlation and similarity between signals, and realizing the self-adaptive learning and prediction of the model;
the main formulas of the decoder are as follows:
;
;
;
;
wherein,representing the output of the layer I encoder,/I>Respectively representing the seasonal component and the trend component after the i-th component of the first layer is separated, the ++>Respectively represent the trend of the ith extraction +.>Is a projection of (2);
the final prediction result is the sum of two refinement decomposition components, and the formula is:
;
wherein,is to transform the depth-transformed seasonal component +.>Projected to the target.
7. A computer storage medium having stored thereon a computer program which, when executed by a processor, implements a short-term wind power prediction method based on spatio-temporal correlation according to any of claims 1-6.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements a short-term spatio-temporal correlation-based wind power prediction method according to any of claims 1-6 when executing the computer program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410235440.9A CN117810997A (en) | 2024-03-01 | 2024-03-01 | Short-term wind power prediction method based on space-time correlation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410235440.9A CN117810997A (en) | 2024-03-01 | 2024-03-01 | Short-term wind power prediction method based on space-time correlation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117810997A true CN117810997A (en) | 2024-04-02 |
Family
ID=90420386
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410235440.9A Pending CN117810997A (en) | 2024-03-01 | 2024-03-01 | Short-term wind power prediction method based on space-time correlation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117810997A (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108535636A (en) * | 2018-05-16 | 2018-09-14 | 武汉大学 | A kind of analog circuit is distributed the neighbouring embedded fault signature extracting method that the victor is a king based on stochastic parameter |
CN113988359A (en) * | 2021-09-08 | 2022-01-28 | 中南大学 | Wind power prediction method and system based on asymmetric Laplace distribution |
CN116845874A (en) * | 2023-07-06 | 2023-10-03 | 固德威技术股份有限公司 | Short-term prediction method and device for power load |
-
2024
- 2024-03-01 CN CN202410235440.9A patent/CN117810997A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108535636A (en) * | 2018-05-16 | 2018-09-14 | 武汉大学 | A kind of analog circuit is distributed the neighbouring embedded fault signature extracting method that the victor is a king based on stochastic parameter |
CN113988359A (en) * | 2021-09-08 | 2022-01-28 | 中南大学 | Wind power prediction method and system based on asymmetric Laplace distribution |
CN116845874A (en) * | 2023-07-06 | 2023-10-03 | 固德威技术股份有限公司 | Short-term prediction method and device for power load |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ma et al. | A hybrid attention-based deep learning approach for wind power prediction | |
CN112200244B (en) | Intelligent detection method for anomaly of aerospace engine based on hierarchical countermeasure training | |
CN110067696B (en) | Wind turbine generator load determining method, device, equipment and readable storage medium | |
CN113822418A (en) | Wind power plant power prediction method, system, device and storage medium | |
Wang et al. | Application of DBN for estimating daily solar radiation on horizontal surfaces in Lhasa, China | |
CN116680105A (en) | Time sequence abnormality detection method based on neighborhood information fusion attention mechanism | |
CN116706907B (en) | Photovoltaic power generation prediction method based on fuzzy reasoning and related equipment | |
CN112559741B (en) | Nuclear power equipment defect record text classification method, system, medium and electronic equipment | |
CN113221992A (en) | Based on L2,1Large-scale data rapid clustering method of norm | |
CN117810997A (en) | Short-term wind power prediction method based on space-time correlation | |
CN110717623A (en) | Photovoltaic power generation power prediction method, device and equipment integrating multiple weather conditions | |
CN115545164A (en) | Photovoltaic power generation power prediction method, system, equipment and medium | |
CN112561180B (en) | Short-term wind speed prediction method and device based on meta-learning, computer equipment and storage medium | |
CN115100599A (en) | Mask transform-based semi-supervised crowd scene abnormality detection method | |
CN113627670A (en) | Ultra-short-term photovoltaic power prediction system based on deep Fourier network | |
CN112529285A (en) | Photovoltaic power generation power prediction method based on similar daily theory and PCA-PSO-BP | |
CN117633456B (en) | Marine wind power weather event identification method and device based on self-adaptive focus loss | |
CN112215383A (en) | Distributed photovoltaic power generation power prediction method and system | |
CN112541554B (en) | Multi-mode process monitoring method and system based on time constraint and nuclear sparse representation | |
CN117828488B (en) | Solar radiance prediction method based on random forest and robust regression | |
CN117933531A (en) | Distributed photovoltaic power generation power prediction system and method | |
CN112712192B (en) | Coal mine gas concentration prediction method combining integrated learning and weighted extreme learning machine | |
CN117609772A (en) | Fault prediction method and system based on nuclear turbine | |
CN115860247A (en) | Method and device for training fan loss power prediction model in extreme weather | |
CN117494896A (en) | Photovoltaic power short-term prediction method and system based on Gaussian mixture model clustering |
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 |