CN119443356A

CN119443356A - Wind power generation prediction method and electronic equipment based on deep learning

Info

Publication number: CN119443356A
Application number: CN202411402244.2A
Authority: CN
Inventors: 马东进; 高英才; 李伟; 胡洪波; 郑晨
Original assignee: CHN Energy Group Science and Technology Research Institute Co Ltd
Current assignee: CHN Energy Group Science and Technology Research Institute Co Ltd
Priority date: 2024-10-09
Filing date: 2024-10-09
Publication date: 2025-02-14

Abstract

The present application provides a wind power prediction method and electronic device based on deep learning, belonging to the field of wind power generation technology, the method comprising: acquiring time series meteorological data; reconstructing the time series meteorological data based on the encoder-decoder module in the trained wind power prediction model to obtain reconstructed data; projecting the reconstructed data to the Legendre polynomial space for feature extraction based on the Legendre polynomial projection module in the trained wind power prediction model to obtain a feature vector; performing spectrum convolution processing on the feature vector based on the Fourier transform module in the trained wind power prediction model to obtain data after spectrum convolution; processing the data after spectrum convolution based on the multi-layer perceptron module in the trained wind power prediction model to obtain wind power. The present application scheme can improve the prediction precision and accuracy of wind power.

Description

Deep learning-based wind power generation power prediction method and electronic equipment

Technical Field

The application relates to the technical field of wind power generation, in particular to a wind power generation power prediction method based on deep learning and electronic equipment.

Background

Currently, wind power generation power prediction is mainly a method based on meteorological data and historical power data. The meteorological data comprise wind power, wind speed, wind direction, temperature, humidity and other data, and the historical power data are used for building a prediction model. The commonly used prediction models comprise models such as an artificial neural network, a support vector machine, regression analysis and the like, the prediction models can correlate meteorological data with historical power data, and a prediction result of wind power generation power is obtained through the trained prediction models.

In the prior art, the wind power generation power prediction method mainly comprises wind power generation power prediction based on a statistical method and wind power generation power prediction based on a physical model. However, these two prediction methods have low prediction accuracy and poor accuracy. Therefore, how to improve the prediction accuracy and precision of the wind power generation power becomes a problem yet to be solved.

Disclosure of Invention

The application aims to provide a wind power generation power prediction method based on deep learning, which can solve the problems of lower prediction precision and poorer accuracy of wind power generation power in the prior art.

In a first aspect, an embodiment of the present application provides a wind power generation power prediction method based on deep learning, where the method includes:

acquiring time sequence meteorological data;

Reconstructing the time-series meteorological data based on an encoder-decoder module in the trained wind power generation power prediction model to obtain reconstructed data;

Based on a Legendre polynomial projection module in the trained wind power generation power prediction model, projecting the reconstructed data to a Legendre polynomial space for feature extraction to obtain feature vectors;

performing spectrum convolution processing on the feature vector based on a Fourier transform module in the trained wind power generation power prediction model to obtain spectrum convolved data;

and processing the data after the spectrum convolution based on the multi-layer perceptron module in the trained wind power generation power prediction model to obtain the wind power generation power.

In a possible implementation manner of the first aspect, after acquiring the time-series meteorological data, the method further includes:

And carrying out average pooling downsampling on the time-series meteorological data to obtain downsampled data, wherein the downsampled data has the following expression:

Wherein, The downsampled data is represented, t represents time, k represents a pooling window size, i represents ordinal numbers, x (t) represents time-series meteorological data, and x (i) represents i time-series meteorological data with t in x (t) replaced.

In a possible implementation manner of the first aspect, reconstructing the time-series meteorological data based on an encoder-decoder module in the trained wind power generation power prediction model to obtain reconstructed data includes:

The downsampled data is encoded into an implicit representation based on an encoder, the specific expression being as follows:

Wherein h (t) represents an implicit representation, encoder represents an encoder, Representing the downsampled data;

decoding the implicit representation into reconstructed data based on a decoder, the specific expression is as follows:

Wherein, Representing the reconstructed data, and the Decoder represents the Decoder.

In one possible implementation of the first aspect, the encoder comprises a number of layers of neural networks and the decoder comprises a number of layers of neural networks, the encoding of the downsampled data into the implicit representation based on the encoder comprises:

The downsampled data is encoded into an implicit representation based on several layers of neural networks, the expression of the output of each layer of neural network of the encoder is as follows:

h_l＝f(W_lh_l-1+b_l)

Wherein h _l represents the output of the first layer neural network of the encoder, l represents the index of each layer neural network, f represents the activation function of the encoder, W _l represents the weight matrix of the first layer of the encoder, h _l-1 represents the output of the first-1 layer neural network of the encoder, and b _l represents the bias vector of the first layer of the encoder;

The expression of the implicit representation of the final output of the encoder is as follows:

h(t)=h_L

Where h (t) represents an implicit representation, L represents the total number of layers of the neural network of the encoder, and h _L represents the output of the last layer of the neural network of the encoder.

In a possible implementation manner of the first aspect, decoding the implicit representation into the reconstructed data based on the decoder includes:

Based on several layers of neural networks, the implicit representation is converted into reconstructed data, and the expression of the output of each layer of neural network of the decoder is as follows:

d_l＝g(W_l′d_l-1+b_l′)

Wherein d _l represents the output of the first layer neural network of the decoder, g represents the activation function of the decoder, W _l 'represents the weight matrix of the first layer of the decoder, d _l-1 represents the output of the first-1 layer neural network of the decoder, and b _l' represents the bias vector of the first layer of the decoder;

the expression of the reconstructed data finally output by the encoder is as follows:

Wherein, Representing reconstructed data, L' represents the total number of layers of the neural network of the decoder, and d _L′ represents the output of the last layer of the neural network of the decoder.

In a possible implementation manner of the first aspect, before the projecting module of the Legendre polynomial in the wind power generation power prediction model after training projects the reconstructed data to the space of the Legendre polynomial to perform feature extraction to obtain the feature vector, the method further includes:

The data normalization module in the wind power generation power prediction model after training is used for carrying out data normalization processing on the reconstructed data to obtain reconstructed data after the data normalization processing, and the expression of the reconstructed data after the data normalization processing is as follows:

Wherein, Represents the reconstructed data after the data normalization process,Represents the reconstructed data, mu ₁ represents the mean value of the reconstructed data, sigma ₁ represents the standard deviation of the reconstructed data.

In a possible implementation manner of the first aspect, the projecting the reconstructed data to the Legendre polynomial space based on the Legendre polynomial projection module in the trained wind power generation power prediction model to perform feature extraction to obtain the feature vector includes:

for each moment, calculating each order Legendre polynomial corresponding to the reconstructed data after the data normalization processing, wherein the expression of each order Legendre polynomial is as follows:

Wherein, Represents the reconstructed data after the data normalization process,Is a 0 th order Legendre polynomial,Is a Legendre polynomial of the 1 st order,Is an nth order Legendre polynomial, n is a natural number;

And weighting and summing the Legendre polynomials of each step according to the corresponding coefficients to obtain a feature vector, wherein the expression of the feature vector is as follows:

Wherein P (t) represents a feature vector, N represents the length of the feature vector, and a _n represents coefficients corresponding to each order Legendre polynomial.

In a possible implementation manner of the first aspect, the performing, based on a fourier transform module in the trained wind power generation power prediction model, a spectral convolution processing on the feature vector to obtain data after the spectral convolution includes:

the eigenvectors are fourier transformed, and the specific expression is as follows:

F[P(t)]=Φ[P(t)]

Wherein F is an intermediate variable, P (t) represents a feature vector, and phi represents Fourier transform;

the feature vector after fourier transformation is subjected to linear transformation, and the specific expression is as follows:

S_f＝W_f·F[P(t)]

Wherein S _f represents a linear transformation, and W _f represents a linear transformation matrix;

The feature vector after the linear transformation is subjected to inverse Fourier transformation, and the specific expression is as follows:

S(t)=Φ^-1(S_f)

where S (t) represents the data after spectral convolution, Φ ^-1 represents the inverse fourier transform.

In a possible implementation manner of the first aspect, performing fourier transformation on the feature vector includes:

Wherein F [ P (t) ] (k) represents the Fourier transform of P (t), P (t) represents the eigenvector, k represents the frequency index, N represents the time index, N represents the length of the eigenvector, e is a natural constant, and i is an imaginary unit;

Performing linear transformation on the eigenvector after Fourier transformation, including:

wherein S _f (k) represents a linear transformation, W _f (k, j) is an element of a kth row and a jth column in the linear transformation matrix, F [ P (t) ] (j) represents a Fourier transformation of P (t), k represents a row of the linear transformation matrix, and j represents a column of the linear transformation matrix;

Performing inverse fourier transform on the feature vector after the linear transformation, including:

where S (t) _n denotes data after spectral convolution.

In one possible implementation manner of the first aspect, the multi-layer perceptron module includes an input layer, a hidden layer and an output layer, and the processing the data after the spectrum convolution based on the trained multi-layer perceptron module to obtain wind power generation power includes:

The data S (t) _n after spectrum convolution is taken as an input vector x at an input layer;

At the hidden layer, the input of the first layer is x _l-1, the output of the first layer is x _l, and the expression of the operation of the first layer is as follows:

z_l＝W_lx_l-1+b_l

Wherein z _l denotes the operation of the first layer, W _l denotes the weight matrix of the first layer, b _l denotes the bias vector of the first layer, and l denotes the hidden layer ordinal number;

Z _l is processed by using an activation function, and the specific expression is as follows:

x_l＝f(z_l)

Where x _l represents the output of the first layer and f represents the activation function;

In the output layer, a full connection layer is adopted, the input of the full connection layer is the output of the last hidden layer, the output of the full connection layer is a prediction result, and the specific expression is as follows:

Wherein, Representing the prediction result, W _l represents the weight matrix of the first layer, and b _l represents the bias vector of the first layer;

carrying out data inverse normalization processing on the prediction result to obtain wind power generation power:

Where y (t) represents the wind power generation power, σ ₂ represents the standard deviation of the prediction result, and μ ₂ represents the average value of the prediction result.

In a possible implementation manner of the first aspect, the method further includes:

the method comprises the steps of training a wind power generation power prediction model according to historical meteorological data and historical wind power generation power data, wherein the wind power generation power prediction model comprises an encoder-decoder module, a data normalization module, a Legendre polynomial projection module, a Fourier transform module and a multi-layer perceptron module, and updating weight parameters of the trained wind power generation power prediction model according to latest meteorological data and latest wind power generation power data in the process of predicting wind power generation power based on the trained wind power generation power prediction model.

In a second aspect, an embodiment of the present application provides a wind power prediction apparatus based on deep learning, including:

The acquisition unit is used for acquiring time sequence meteorological data;

the reconstruction unit is used for reconstructing the time sequence meteorological data based on an encoder-decoder module in the trained wind power generation power prediction model to obtain reconstruction data;

the feature extraction unit is used for projecting the reconstructed data to the Legendre polynomial space to perform feature extraction based on the Legendre polynomial projection module in the trained wind power generation power prediction model to obtain feature vectors;

the frequency spectrum convolution unit is used for carrying out frequency spectrum convolution processing on the feature vector based on a Fourier transform module in the trained wind power generation power prediction model to obtain data after frequency spectrum convolution;

the output unit is used for processing the data after the spectrum convolution based on the multi-layer perceptron module in the trained wind power generation power prediction model to obtain wind power generation power.

In a third aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor executes the computer program to implement any one of the deep learning-based wind power generation power prediction methods of the first aspect.

In a fourth aspect, an embodiment of the present application provides a computer readable storage medium storing a computer program, where the computer program when executed by a processor implements any one of the deep learning-based wind power prediction methods of the first aspect.

In a fifth aspect, embodiments of the present application provide a computer program product for, when run on an electronic device, causing the electronic device to perform any one of the deep learning based wind power generation power prediction methods of the first aspect above.

According to the application, firstly, time sequence meteorological data is rebuilt through an encoder-decoder module, secondly, characteristics of the rebuilt data are extracted through a Legendre polynomial projection module to obtain characteristic vectors, then, the characteristic vectors are subjected to spectrum convolution processing through a Fourier transform module, and finally, the data subjected to spectrum convolution are processed through a multi-layer perceptron module to obtain wind power generation power. The scheme of the application can improve the prediction precision and accuracy of wind power generation power, and has strong usability and practicability.

Additional features and advantages of the application will be set forth in the detailed description which follows.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic step diagram of a wind power generation power prediction method based on deep learning according to an embodiment of the present application;

FIG. 2 is a schematic diagram of overall steps of a deep learning-based wind power prediction method according to an embodiment of the present application;

FIG. 3 is a schematic structural diagram of a wind power generation power prediction device based on deep learning according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

It should be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

It is also to be understood that the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

It should be further understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.

As used in this specification and the appended claims, the term "if" may be interpreted as "when..once" or "in response to a determination" or "in response to detection" depending on the context. Similarly, the phrase "if a determination" or "if a [ described condition or event ] is detected" may be interpreted in the context of meaning "upon determination" or "in response to determination" or "upon detection of a [ described condition or event ]" or "in response to detection of a [ described condition or event ]".

In addition, in the description of the present application, the terms "first," "second," "third," etc. are used merely to distinguish between descriptions and should not be construed as indicating or implying relative importance.

Reference in the specification to "one embodiment" or "some embodiments" or the like means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the application. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in some other embodiments," "in other embodiments," and the like in the specification are not necessarily all referring to the same embodiment, but mean "one or more, but not all, embodiments" unless otherwise expressly specified. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless expressly specified otherwise.

With the increase of global energy crisis and the increase of environmental awareness, renewable energy sources (such as wind energy and solar energy) and the like are widely paid attention to and utilized. Wind power generation is one of the main forms of wind energy utilization, and prediction of the generated power has important significance for stable operation of a power system, reasonable scheduling of energy sources and effective trading of markets.

The power generated by a wind power station is affected by a number of factors, such as wind power, wind speed, wind direction, temperature, humidity, etc. The wind power generation power prediction is to input a weather element prediction conclusion into a power generation power model, calculate the power generation capacity of a wind power station, generate and transmit power in a grid-connected mode according to the power generation capacity, and finally finish transaction in an electric market. For distributed power generation systems, accurate prediction of power generation capacity is important. How to accurately predict power output, better plan and schedule power generation behavior, use green electricity to the greatest extent, and is one of key technologies for ensuring safe and stable operation of a power system.

In the prior art, the wind power generation power prediction method mainly comprises a wind power generation power prediction method based on a statistical method and a wind power generation power prediction method based on a physical model. The wind power generation power prediction method based on the statistical method mainly relies on historical data and a statistical model to predict the future wind power generation power of a wind farm, and the prediction model is established by analyzing modes, trends and correlations in the historical data. Common statistical methods include regression analysis, time series analysis, gray prediction, and the like. The models can capture the influence of factors such as weather, seasons, time and the like on the power generation power of the wind farm, and forecast the influence.

The wind power generation power prediction method based on the physical model predicts the power generation of the wind power plant according to the physical characteristics and the operation principle of the wind power plant and combines meteorological data and environmental factors. Such methods often require extensive knowledge of the components, structure, and manner of operation of the wind farm. The common physical models comprise a wind power model, a temperature model, an efficiency model and the like, and the models can simulate the running conditions of the wind power plant under different weather and environmental conditions and predict the power generation power of the wind power plant according to the running conditions.

The problems of the existing wind power generation power prediction method mainly come from the characteristics of fluctuation, intermittence and randomness of wind energy, the limitation of the prediction technology and the complexity of practical application. The method mainly has the problems that firstly, the wind speed prediction accuracy is deficient, the existing prediction technology is used for predicting the average wind speed of the wind power plant, and the complex influence of the topography and the topography of the wind power plant on the wind speed is not fully considered. The simplified prediction method causes deviation between the predicted wind speed and the wind speed at the actual wind turbine generator set, thereby affecting the accuracy of power prediction.

Secondly, the power conversion relation is simple, namely, in the process of converting the predicted wind speed into wind power, the simple conversion relation is generally adopted, and the influences of practical factors such as towers, wind shearing differences and the like are not fully considered. This simplified scaling method results in insufficient prediction accuracy of wind power. Thirdly, the combination algorithm is insufficient, namely when wind power generation prediction is carried out by utilizing the combination algorithm, only simple weight combination is carried out on the prediction result of each algorithm, and the combination weight is optimized by not fully combining the defects of the algorithm.

Aiming at the defects, the embodiment of the application provides a wind power generation power prediction method based on deep learning, which comprises the steps of firstly reconstructing time sequence meteorological data through an encoder-decoder module, secondly extracting features of reconstructed data through a Legendre polynomial projection module to obtain feature vectors, then carrying out spectrum convolution processing on the feature vectors through a Fourier transform module, and finally processing the data subjected to spectrum convolution through a multi-layer perceptron module to obtain wind power generation power. The scheme of the application can improve the prediction precision and accuracy of wind power generation power, and has strong usability and practicability.

The following describes specific procedures performed by the present application by way of specific examples.

Referring to fig. 1, fig. 1 is a schematic diagram illustrating steps of a wind power generation power prediction method based on deep learning according to an embodiment of the present application. As shown in fig. 1, the method may include the steps of:

s101, acquiring time sequence meteorological data.

In one embodiment, meteorological data such as wind power, wind speed, wind direction, temperature, humidity, etc. of a wind power station over a preset period of time may be collected over time. The preset time period is a future time period, and may be determined according to a specific situation in an actual application scenario, which is not specifically limited herein.

According to one embodiment of the present application, after acquiring the time-series meteorological data, the method may further comprise the steps of:

It should be noted that wind power generation time series meteorological data generally has a significant long-short-term dependency relationship, and it is difficult for a traditional prediction model to capture the dependency relationship of a long time and a short time at the same time.

S102, reconstructing time-series meteorological data based on an encoder-decoder module in the trained wind power generation power prediction model to obtain reconstructed data.

According to an embodiment of the present application, reconstructing time-series meteorological data based on an encoder-decoder module in a trained wind power generation power prediction model to obtain reconstructed data may include the steps of:

Wherein h (t) represents an implicit representation, encoder represents an encoder, Representing the downsampled data.

According to one embodiment of the application, the encoder comprises a plurality of layers of neural networks and the decoder comprises a plurality of layers of neural networks, and the encoding of the downsampled data into the implicit representation based on the encoder may comprise the steps of:

h_l＝f(W_lh_l-1+b_l)

Where h _l denotes the output of the first layer neural network of the encoder, l denotes the index of each layer neural network, f denotes the activation function of the encoder, W _l denotes the weight matrix of the first layer of the encoder, h _l-1 denotes the output of the first-1 layer neural network of the encoder, b _l denotes the bias vector of the first layer of the encoder, the weight matrix and the bias vector can be obtained by learning.

h(t)=h_L

According to one embodiment of the application, decoding the implicit representation into reconstructed data based on the decoder may comprise the steps of:

d_l＝g(W_l′d_l-1+b_l′)

Where d _l denotes the output of the first layer neural network of the decoder, g denotes the activation function of the decoder, W _l 'denotes the weight matrix of the first layer of the decoder, s _l-1 denotes the output of the first-1 layer neural network of the decoder, b _l' denotes the bias vector of the first layer of the decoder, the weight matrix and the bias vector being obtainable by learning.

In the embodiment, the encoder-decoder module can effectively capture the long-term and short-term dependency relationship of the time-series meteorological data, and improve the accuracy of wind power generation power prediction.

It should be noted that, the original time series meteorological data has higher dimensionality and more redundant information, and direct modeling may cause problems of higher complexity of the model or over fitting.

And S103, based on a Legendre polynomial projection module in the trained wind power generation power prediction model, projecting the reconstructed data to a Legendre polynomial space to perform feature extraction, and obtaining feature vectors.

It should be noted that wind power generation data may have different dimensions and ranges, and direct modeling using these data may result in a model with a slow convergence rate, a low prediction accuracy, and even unstable values. According to one embodiment of the present application, before the reconstructed data is projected to the Legendre polynomial space based on the Legendre polynomial projection module in the trained wind power generation power prediction model to perform feature extraction to obtain the feature vector, the method may further include the following steps:

In this embodiment, the data normalization module is used to perform normalization processing on the reconstructed data, so that the dimension and the range of the data are the same, and the numerical stability and the training efficiency of the model can be improved.

According to one embodiment of the present application, based on the Legendre polynomial projection module in the trained wind power generation power prediction model, the reconstructed data is projected to the Legendre polynomial space to perform feature extraction to obtain feature vectors, which may include the following steps:

Wherein, Represents the reconstructed data after the data normalization process,Is a 0 th order Legendre polynomial,Is a Legendre polynomial of the 1 st order,Is an nth order Legendre polynomial, n is a natural number.

Wherein P (t) represents a feature vector, N represents the length of the feature vector, a _n represents coefficients corresponding to the Legendre polynomials of each order, and the coefficients can be obtained by learning.

In the embodiment, the Legendre polynomial projection module is adopted to project the reconstructed data subjected to the data normalization processing into the Legendre polynomial space, so that key features are extracted, the redundancy of the data is reduced, and the generalization capability and the prediction performance of the model are improved.

It should be noted that wind power generation data generally has significant periodicity and frequency characteristics, and it is difficult for a conventional time domain model to effectively capture these characteristics.

S104, carrying out spectrum convolution processing on the feature vector based on a Fourier transform module in the trained wind power generation power prediction model to obtain data after spectrum convolution.

According to an embodiment of the present application, the performing spectral convolution processing on the feature vector based on the fourier transform module in the trained wind power generation power prediction model to obtain the data after spectral convolution may include the following steps:

F[P(t)]=Φ[P(t)]

Wherein F is an intermediate variable, P (t) is a eigenvector, phi is a Fourier transform, which can be FFT (fast Fourier transform ) with a relatively high computation speed, and is used for transforming a time domain signal into a frequency domain signal, P (t) is a time domain signal, and F [ P (t) ] is a frequency domain signal.

S_f＝W_f·F[P(t)]

Wherein S _f represents a linear transformation, and W _f represents a linear transformation matrix, which can be obtained by learning, and parameters of which can be optimized according to back propagation of historical meteorological data and historical wind power generation power data.

S(t)=φ^-1(S_f)

Where S (t) represents the data after spectral convolution, phi ^-1 represents the inverse fourier transform, which may be IFFT (INVERSE FAST Fourier Transform ), for transforming the frequency domain signal back to the time domain signal, S _f is the frequency domain signal, and S (t) is the time domain signal.

According to one embodiment of the application, the specific formula for fourier transforming the eigenvectors is as follows:

Wherein F [ P (t) ] (k) represents the Fourier transform of P (t), P (t) represents the eigenvector, k represents the frequency index, N represents the time index, N represents the length of the eigenvector, e is a natural constant, and i is an imaginary unit.

The specific formula for performing linear transformation on the eigenvectors after fourier transformation is as follows:

where S _f (k) represents a linear transformation, W _f (k, j) is an element of the kth row and the jth column in the linear transformation matrix, F [ P (t) ] (j) represents a fourier transformation of P (t), k represents a row of the linear transformation matrix, and j represents a column of the linear transformation matrix.

The specific formula for performing inverse fourier transform on the feature vector after linear transformation is as follows:

where S (t) _n denotes data after spectral convolution.

In the embodiment, the Fourier transform module is adopted to capture the periodicity and the frequency characteristics in the data through frequency domain characteristic extraction, so that the processing capacity of the model on the periodicity change is enhanced, and the prediction accuracy is improved.

It should be noted that there are complex nonlinear relations in wind power generation data, and it is difficult to effectively model these relations by using a simple linear model.

S105, processing the data after the spectrum convolution based on the multi-layer perceptron module in the trained wind power generation power prediction model to obtain wind power generation power.

In one embodiment, the data after the spectrum convolution may be input to an MLP (Multilayer Perceptron, multi-layer perceptron) module to perform final prediction, so as to obtain a final prediction result, that is, wind power, where a specific expression is as follows:

Wherein, The wind power generation power is represented, the MLP is data processing of the multi-layer perceptron module, and S (t) is data after spectrum convolution.

According to one embodiment of the present application, a multi-layer perceptron module includes an input layer, a hidden layer, and an output layer. The data after spectrum convolution is processed based on the trained multi-layer perceptron module to obtain wind power generation power, and the method can comprise the following steps:

At the input layer, the spectrally convolved data S (t) _n is taken as the input vector x. At the hidden layer, the input of the first layer is x _l-1, the output of the first layer is x _l, and the expression of the operation of the first layer is as follows:

z_l＝W_lx_l-1+b_l

Wherein z _l represents the operation of the first layer, W _l represents the weight matrix of the first layer, b _l represents the bias vector of the first layer, and l represents the hidden layer ordinal number, and the weight matrix and bias vector can be obtained by learning.

x_l＝f(z_l)

where x _l represents the output of the first layer and f represents the activation function.

Wherein, The predicted result is represented by W _l, the weight matrix of the first layer is represented by b _l, the bias vector of the first layer is represented by b _l, and the weight matrix and the bias vector can be obtained through learning.

It should be noted that, the prediction model needs to maintain consistency between the preprocessed data and the output result, so as to ensure that no deviation occurs in the transmission and processing processes of the data in the model. In one embodiment, the prediction result is subjected to data inverse normalization processing to obtain wind power generation power:

Where y (t) represents the wind power generation power, σ ₂ represents the standard deviation of the prediction result, and μ ₂ represents the average value of the prediction result. The data normalization module and the data inverse normalization module ensure the consistency of data in the input and output stages, so that the data processed by the model are consistent in dimension and range, and the accuracy of a prediction result is ensured.

In this embodiment, the multi-layer perceptron module performs nonlinear modeling on the extracted features through multi-layer nonlinear mapping, captures a complex nonlinear relationship, and improves the prediction performance of the model.

According to one embodiment of the application, the method may further comprise the steps of:

Training a wind power generation power prediction model according to historical meteorological data and historical wind power generation power data, wherein the wind power generation power prediction model comprises an encoder-decoder module, a data normalization module, a Legendre polynomial projection module, a Fourier transform module and a multi-layer perceptron module. In the process of predicting wind power generation power based on the trained wind power generation power prediction model, updating the weight parameters of the trained wind power generation power prediction model according to the latest meteorological data and the latest wind power generation power data.

The weight parameters of the wind power generation power prediction model comprise weight matrixes and offset vectors in an encoder-decoder module, corresponding coefficients of each order Legendre polynomial in a Legendre polynomial projection module, parameters of a linear transformation matrix in a Fourier transformation module and weight matrixes and offset vectors in a multi-layer perceptron module.

It should be noted that, because the data normalization module does not include the weight parameter that needs to be obtained through training, the data normalization module does not need to be trained, and other modules can be trained. The training goal is to minimize the mean square error of wind power generation power, the training loss is minimum, and the mean square error is minimum. The training loss is calculated according to a loss function, and the expression of the loss function is as follows:

Wherein L represents a loss function, N represents a length of the feature vector, i represents an ordinal number, y _i represents an actual value of wind power generation power, A predicted value of the wind power generation power is indicated.

It should be noted that wind power generation predictions may require the incorporation of multi-source data (e.g., meteorological data, historical wind power data, etc.), how to effectively fuse these data and enhance the characterization is a challenge. Through modularized design, each module (such as an encoder-decoder module, a Legendre polynomial projection module and a Fourier transform module) respectively processes different types of characteristic information, and finally comprehensive modeling is carried out in a multi-layer perceptron module, so that effective fusion and characteristic enhancement of multi-source data are realized.

It should be noted that the training efficiency and convergence speed of the complex model are important considerations in practical applications, and it may be difficult to achieve a better effect in a reasonable time by using the conventional model. The training stability is improved through data normalization, the complexity of the model is reduced through modularized design and feature extraction (such as a Legendre polynomial projection module and a Fourier transform module), and the training efficiency and the convergence speed of the model are improved.

According to the embodiment of the application, a deep learning model is trained through historical data, the model is used for predicting wind power generation power in a future period, and the latest data is used for updating model parameters, so that the accuracy of prediction is improved. Firstly, a coder-decoder module based on a transducer focuses on important parts in input time sequence meteorological data, secondly, the data normalization module is used for carrying out standardized processing on the output of the coder to avoid the problems of different dimensions, large difference of numerical ranges and the like of wind power generation data, thirdly, the Legendre polynomial projection module is used for carrying out feature extraction to solve the problems of retaining and utilizing history information, then, the dimension is reduced by a method combining Fourier transformation and linear transformation matrix approximation objective function, the influence of noise signals in the time sequence meteorological data is furthest reduced, the problem of excessive extraction is relieved, finally, the multi-layer perceptron module is used for predicting, and the predicted result is output after inverse normalization, and finally, the wind power generation power is obtained through prediction.

It should be noted that, the key of the encoder-decoder module is to be able to effectively capture the long-short-term dependency of the input time sequence, and enhance the modeling capability of the complex time sequence data. The key of the data normalization module and the data inverse normalization module is to normalize data processing, ensure the numerical stability of input data and improve the efficiency and accuracy of model training and prediction. The Legendre polynomial projection module is characterized in that time sequence data is projected into a polynomial space to perform feature extraction, so that the expression capacity of features is enhanced, and the prediction performance of a model is improved. The key point of the Fourier transform module is that periodicity and frequency characteristics in the data are captured through frequency domain characteristic extraction and processing, and the prediction accuracy of the model is improved. The multi-layer perceptron module is characterized in that the extracted characteristics are effectively modeled through multi-layer nonlinear mapping, and finally a high-precision prediction result, namely wind power generation power, is output.

Referring to fig. 2, fig. 2 is a schematic diagram illustrating overall steps of a wind power generation power prediction method based on deep learning according to an embodiment of the present application. As shown in fig. 2, the method may include the steps of:

S201, pre-training a wind power generation power prediction model according to historical data, and updating model parameters according to the latest data in the process of predicting wind power generation power based on the trained wind power generation power prediction model.

The historical data are historical meteorological data and historical wind power generation power data, the latest data are latest meteorological data and latest wind power generation power data, the updated model parameters are weight parameters of an updated wind power generation power prediction model, and the wind power generation power prediction model comprises an encoder-decoder module, a data normalization module, a Legendre polynomial projection module, a Fourier transform module and a multi-layer perceptron module.

S202, acquiring time-series meteorological data x (t);

S203, carrying out average pooling downsampling on time sequence meteorological data x (t) to obtain downsampled data

S204, reconstructing the time-series meteorological data x (t) based on the encoder-decoder module to obtain reconstructed data

S205, reconstructing data based on the data normalization modulePerforming data normalization processing to obtain reconstructed data after the data normalization processing

S206, projecting the reconstructed data to a Legendre polynomial space based on a Legendre polynomial projection module to perform feature extraction to obtain a feature vector P (t);

s207, performing spectrum convolution processing on the feature vector based on the Fourier transform module to obtain spectrum convolved data S (t);

S208, processing the data S (t) after spectrum convolution based on a multi-layer perceptron (MLP) module, and outputting wind power generation power Finally, a prediction result of wind power generation power is obtained.

It should be noted that, the specific implementation principles of steps S201 to S208 are described in the above embodiments, and are not described herein.

According to the wind power generation power prediction method based on deep learning, firstly, time sequence meteorological data are rebuilt through an encoder-decoder module, secondly, features of the rebuilt data are extracted through a Legendre polynomial projection module to obtain feature vectors, then, the feature vectors are subjected to spectrum convolution processing through a Fourier transform module, and finally, the data subjected to spectrum convolution are processed through a multi-layer perceptron module to obtain wind power generation power. The scheme of the embodiment of the application can improve the prediction precision and accuracy of the wind power generation power, and has stronger usability and practicability.

The deep learning model provided by the embodiment of the application can be used for predicting the wind power generation power in a future period by integrating the influence of various factors such as wind power, wind speed, wind direction, temperature, humidity and the like of a wind power station. The deep learning model has the following technical performances that the prediction precision exceeds the level of similar models, accurate wind power generation power prediction data can be provided, the deep learning model has the characteristics of stable performance and high detection precision, can meet the requirements of long-time continuous operation, is suitable for wind power generation power prediction of various wind power plants, and has wide applicability.

In addition, the scheme of the embodiment of the application has the advantages that the encoder-decoder module is used for capturing the long-term and short-term dependency relationship of the input time sequence, and the time sequence characteristic of wind power generation data can be effectively processed. The data normalization module is used for normalizing input data, eliminating dimension difference and improving convergence speed and prediction accuracy of the model. The Legendre polynomial projection module is used for projecting data into a polynomial space to perform feature extraction, and feature representation capacity is improved, so that prediction performance of the model is enhanced. The Fourier transform module captures the periodicity and frequency characteristics in the input data through frequency domain characteristic extraction and processing, and improves the prediction precision. The multi-layer perceptron module is used for nonlinear mapping and final wind power generation power prediction, and can effectively model the extracted features.

And secondly, the method has good generalization capability, namely, through an encoder-decoder module and a Legendre polynomial projection module, effective information can be extracted from different time scales and frequency domain characteristics, and the method has good generalization capability and is suitable for different types of wind power generation data. And thirdly, the interpretability and the feature extraction capability are strong, namely, the Legendre polynomial projection module provides good feature extraction and interpretation capability in a polynomial expansion mode, and is helpful for understanding key features in input data. The fourier transform module introduces frequency domain features that provide the model with the ability to interpret the data periodicity and frequency features.

It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic, and should not limit the implementation process of the embodiment of the present application.

Corresponding to the method of the above embodiment, fig. 3 is a schematic structural diagram of a wind power prediction device based on deep learning according to an embodiment of the present application. For convenience of explanation, only portions relevant to the embodiments of the present application are shown.

Referring to fig. 3, the apparatus includes:

an acquisition unit 301 for acquiring time-series meteorological data;

A reconstruction unit 302, configured to reconstruct time-series meteorological data based on an encoder-decoder module in the trained wind power generation power prediction model, so as to obtain reconstructed data;

The feature extraction unit 303 is configured to project the reconstructed data to a Legendre polynomial space based on a Legendre polynomial projection module in the trained wind power generation power prediction model to perform feature extraction, so as to obtain a feature vector;

the spectrum convolution unit 304 is configured to perform spectrum convolution processing on the feature vector based on a fourier transform module in the trained wind power generation power prediction model, so as to obtain data after spectrum convolution;

And the output unit 305 is used for processing the data after the spectrum convolution based on the multi-layer perceptron module in the trained wind power generation power prediction model to obtain wind power generation power.

It should be noted that, because the content of information interaction and execution process between the above devices/units is based on the same concept as the method embodiment of the present application, specific functions and technical effects thereof may be referred to in the method embodiment section, and will not be described herein.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, the specific names of the functional units and modules are only for distinguishing from each other, and are not used for limiting the protection scope of the present application. The specific working process of the units and modules in the above system may refer to the corresponding process in the foregoing method embodiment, which is not described herein again.

Fig. 4 is a schematic structural diagram of an electronic device 4 according to an embodiment of the present application. As shown in fig. 4, the electronic device 4 of this embodiment may comprise at least one processor 401 (only one shown in fig. 4), a memory 403 and a computer program 402 stored in the memory 403 and executable on the at least one processor 401, the processor 401 executing the computer program 402 to carry out the steps in the above-described method embodiments.

The electronic device 4 may be a desktop computer, a notebook computer, a palm computer, a mobile phone, or other computing devices. The electronic device 4 may include, but is not limited to, a processor 401, a memory 403. It will be appreciated by those skilled in the art that fig. 4 is merely an example of the electronic device 4 and is not meant to be limiting of the electronic device 4, and may include more or fewer components than shown, or may combine certain components, or different components, such as may also include input-output devices, network access devices, etc.

The Processor 401 may be a central processing unit (Central Processing Unit, CPU), but the Processor 401 may also be other general purpose processors, digital signal processors (DIGITAL SIGNAL Processor, DSP), application SPECIFIC INTEGRATED Circuit (ASIC), field-Programmable gate array (Field-Programmable GATE ARRAY, FPGA) or other Programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory 403 may in some embodiments be an internal storage unit of the electronic device 4, such as a hard disk or a memory of the electronic device 4. The memory 403 may also be an external storage device of the electronic device 4 in other embodiments, such as a plug-in hard disk provided on the electronic device 4, a smart memory card (SMART MEDIA CARD, SMC), a Secure Digital card (SD), a flash memory card (FLASH CARD), etc. Further, the memory 403 may also include both an internal storage unit and an external storage device of the electronic device 4. The memory 403 is used to store an operating system, application programs, boot Loader (Boot Loader), data, and other programs and the like, such as program codes of computer programs and the like. The memory 403 may also be used to temporarily store data that has been output or is to be output.

The integrated units described above, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the present application may be implemented by a computer program, which may be stored in a computer readable storage medium, for instructing a relevant hardware to implement all or part of the procedure in the method of the above embodiment, and which when executed by a processor may implement the steps as described above for the method embodiment. Wherein the computer program comprises computer program code which may be in the form of source code, object code, executable files or in some intermediate form, etc. The computer readable storage medium may include at least any entity or device capable of carrying computer program code to a computing device/electronic apparatus, a recording medium, a computer Memory, a Read-Only Memory (ROM), a random-access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, and a software distribution medium such as a U-disk, a removable hard disk, a magnetic or optical disk, etc. In some jurisdictions, computer-readable storage media may not be electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.

The embodiments of the present application also provide a computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the respective method embodiments described above.

Embodiments of the present application provide a computer program product for causing an electronic device to perform the steps of the method embodiments described above when the computer program product is run on the electronic device.

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus/electronic device and method may be implemented in other manners. The above-described apparatus/electronic device embodiments are merely illustrative, the above-described division of modules or units is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple units, components may be combined or may be integrated into another system, and some features may be omitted from execution. Alternatively, the indirect coupling, direct coupling, or communication connection between the illustrated and discussed may be through some interfaces, devices, or units, which may be electrical, mechanical, or other form.

The units described above as separate components may or may not be physically separate, and components shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

The foregoing embodiments are merely for illustrating the technical solution of the present application, but not for limiting the same, and although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solution described in the foregoing embodiments may be modified or substituted for some of the technical features thereof, and that such modifications or substitutions do not depart from the spirit and scope of the technical solution of the embodiments of the present application and are intended to be included in the scope of the present application.

Claims

1. A wind power generation power prediction method based on deep learning, the method comprising:

acquiring time sequence meteorological data;

2. The deep learning-based wind power generation power prediction method of claim 1, wherein after acquiring the time-series meteorological data, the method further comprises:

3. The deep learning-based wind power generation power prediction method of claim 2, wherein reconstructing the time-series meteorological data based on an encoder-decoder module in a trained wind power generation power prediction model to obtain reconstructed data comprises:

decoding the implicit representation into reconstructed data based on a decoder, the specific expression being as follows:

4. The deep learning based wind power generation prediction method of claim 3, wherein the encoder comprises a plurality of layers of neural networks and the decoder comprises a plurality of layers of neural networks, wherein the encoding the downsampled data into the implicit representation based on the encoder comprises:

h_l＝f(W_lh_l-1+b_l)

Wherein h represents the output of the first layer neural network of the encoder, l represents the index of each layer neural network, f represents the activation function of the encoder, W _l represents the weight matrix of the first layer of the encoder, h _l-1 represents the output of the first-1 layer neural network of the encoder, and b _l represents the bias vector of the first layer of the encoder;

h(t)=h_L

5. A depth learning based wind power prediction method according to claim 3, characterized in that decoding the implicit representation into reconstructed data based on a decoder comprises:

Converting the implicit representation into the reconstructed data based on several layers of neural networks, the expression of the output of each layer of neural network of the decoder is as follows:

d_l＝g(W_l′d_l-1+b′_l)

Wherein d _l represents the output of the first layer neural network of the decoder, g represents the activation function of the decoder, W _l 'represents the weight matrix of the first layer of the decoder, d _l-1 represents the output of the first-1 layer neural network of the decoder, and b' _l represents the bias vector of the first layer of the decoder;

6. The deep learning-based wind power generation power prediction method according to claim 3, wherein before the Legendre polynomial projection module in the trained wind power generation power prediction model projects the reconstructed data into the Legendre polynomial space for feature extraction to obtain feature vectors, the method further comprises:

And carrying out data normalization processing on the reconstructed data based on a data normalization module in the trained wind power generation power prediction model to obtain reconstructed data after the data normalization processing, wherein the expression of the reconstructed data after the data normalization processing is as follows:

7. The deep learning-based wind power generation power prediction method of claim 6, wherein the feature extraction is performed by projecting the reconstructed data to a Legendre polynomial space based on a Legendre polynomial projection module in a trained wind power generation power prediction model to obtain feature vectors, and the method comprises the following steps:

And weighting and summing the Legendre polynomials of each step according to the corresponding coefficients to obtain feature vectors, wherein the expression of the feature vectors is as follows:

8. The deep learning-based wind power generation power prediction method according to claim 7, wherein the performing the spectral convolution processing on the feature vector based on the fourier transform module in the trained wind power generation power prediction model to obtain the data after the spectral convolution comprises:

and carrying out Fourier transform on the characteristic vector, wherein the specific expression is as follows:

F[P(t)]=Φ[P(t)]

S_f＝W_f·F[P(t)]

S(t)=Φ^-1(S_f)

9. The deep learning-based wind power generation power prediction method of claim 8, wherein fourier transforming the feature vector comprises:

where S (t) _n denotes data after spectral convolution.

10. The deep learning-based wind power prediction method as set forth in claim 8, wherein the multi-layer perceptron module includes an input layer, a hidden layer, and an output layer, and wherein the processing of the spectrum convolved data based on the trained multi-layer perceptron module to obtain wind power comprises:

z_l＝W_lx_l-1+b_l

x_l＝f(z_l)

11. The deep learning based wind power generation prediction method of any one of claims 1-10, further comprising:

Training the wind power generation power prediction model according to historical meteorological data and historical wind power generation power data, wherein the wind power generation power prediction model comprises an encoder-decoder module, a data normalization module, a Legendre polynomial projection module, a Fourier transform module and a multi-layer perceptron module;

In the process of predicting wind power generation power based on the trained wind power generation power prediction model, updating the weight parameters of the trained wind power generation power prediction model according to the latest meteorological data and the latest wind power generation power data.

12. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the deep learning based wind power prediction method of any one of claims 1-11 when the computer program is executed.

13. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the deep learning based wind power generation power prediction method of any one of claims 1-11.