CN112348255A - Ultra-short-term wind power prediction method based on wavelet time-frequency imaging - Google Patents

Ultra-short-term wind power prediction method based on wavelet time-frequency imaging Download PDF

Info

Publication number
CN112348255A
CN112348255A CN202011232819.2A CN202011232819A CN112348255A CN 112348255 A CN112348255 A CN 112348255A CN 202011232819 A CN202011232819 A CN 202011232819A CN 112348255 A CN112348255 A CN 112348255A
Authority
CN
China
Prior art keywords
time
layer
frequency
input data
wind power
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.)
Granted
Application number
CN202011232819.2A
Other languages
Chinese (zh)
Other versions
CN112348255B (en
Inventor
何洪英
罗滇生
高思远
蒋宇翔
尹希浩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hunan University
Original Assignee
Hunan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hunan University filed Critical Hunan University
Priority to CN202011232819.2A priority Critical patent/CN112348255B/en
Publication of CN112348255A publication Critical patent/CN112348255A/en
Application granted granted Critical
Publication of CN112348255B publication Critical patent/CN112348255B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/148—Wavelet transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Mathematics (AREA)
  • Marketing (AREA)
  • Computing Systems (AREA)
  • Mathematical Optimization (AREA)
  • General Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Operations Research (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an ultra-short-term wind power prediction method based on wavelet time-frequency imaging, which comprises the steps of obtaining an original input data set, and preprocessing input data of the original input data set; performing wavelet time-frequency analysis on the processed input data, extracting time domain characteristics and frequency domain characteristics of a historical time sequence, and generating a corresponding time-frequency graph; building a neural network prediction model according to the characteristics of the time-frequency diagram, the size of an original input data set and the conditions of a hardware platform; and adjusting parameters of the neural network prediction model according to the image feature extraction result, the training error result and the final prediction result to obtain an optimal prediction result. The method can fully utilize the time-frequency information of the time sequence on one hand and the excellent characteristic of two-dimensional convolution on the other hand, and realizes accurate ultra-short-term prediction of the time sequence of the wind power.

Description

Ultra-short-term wind power prediction method based on wavelet time-frequency imaging
Technical Field
The invention relates to the technical field of time sequence prediction, in particular to an ultrashort-term wind power prediction method based on wavelet time-frequency imaging.
Background
The wavelet transform is proposed by the french physicist Moler in the 80 th 20 th century, is a time-scale (time-frequency) analysis method of signals, and is a time-frequency localization analysis method with a fixed window size, a changeable shape and changeable time window and frequency window.
The resolution of the traditional Fourier transform in time domain and frequency domain is fixed, if the time window is too long, a plurality of periods of the high-frequency signal are included in one time window, and the average value of the plurality of periods is finally output, so that the local characteristics of the high-frequency signal cannot be focused; if the time window is too short, although local features of the high frequency signal may be well focused, features of the low frequency signal may not be detected within one cycle.
The wavelet analysis has good localization property, so that the wavelet analysis has good signal local feature characterization capability in both time domain and frequency domain, and overcomes the inherent defect of single resolution of the traditional Fourier transform.
Convolutional Neural Networks (CNN) is one of the most important methods of deep learning and is a popular research direction in the field of image recognition and speech analysis. The weight sharing network structure is similar to a biological neural network, and the complexity and the weight of a network model can be reduced.
If the network inputs a multi-dimensional image, the image can be directly put into the network, so that the complex processes of feature extraction and data reconstruction in the traditional recognition algorithm can be avoided. As a multi-layered perceptron designed specifically for recognizing two-dimensional shapes, a convolutional neural network has certain processing advantages with respect to the high degree of invariance to translation, scaling, tilting, or deformation of pictures.
Ultra-short term prediction of time series occurs in various engineering applications. At present, the ultra-short term prediction of time series mainly has two problems: on one hand, the signal has certain nonlinearity and non-stationarity, the time-frequency local property is just the most important characteristic of the non-stationary signal, the existing prediction method cannot simultaneously utilize the time domain characteristics and the frequency domain characteristics of the time sequence, and most of the time sequences belong to the non-stationary signal, so that a certain promotion space is provided in the aspect of simultaneously extracting the time-frequency characteristics of the time sequence; on the other hand, the existing ultra-short term prediction models have certain limitations, for example, the improvement of the precision of the prediction result is restricted by the unstable result of the continuous prediction method, the difficulty in estimating the noise amount by using the Kalman filtering method, the lack of sparsity of the LS-SVM model and the like, so that the final prediction result is not accurate enough.
Disclosure of Invention
The invention mainly aims to provide an ultra-short-term wind power prediction method based on wavelet time-frequency imaging, and aims to solve the problem that the prediction result of the existing ultra-short-term prediction model of a time sequence is not accurate enough.
In order to achieve the purpose, the ultrashort-term wind power prediction method based on wavelet time-frequency imaging provided by the invention comprises the following steps:
acquiring historical wind power data as an original input data set, and preprocessing input data of the original input data set;
performing wavelet time-frequency analysis on the processed input data, acquiring wavelet coefficients, and generating a corresponding two-dimensional time-frequency graph according to time domain characteristics and frequency domain characteristics;
building a neural network prediction model for processing the time-frequency diagram according to the characteristics of the time-frequency diagram, the size of the original input data set and the conditions of a hardware platform;
obtaining an image feature extraction result, a training error result and a final prediction result according to an original input data set and the neural network prediction model, and adjusting parameters of the neural network prediction model according to the image feature extraction result, the training error result and the final prediction result to obtain an optimal wind power ultra-short term prediction result.
Preferably, the step of acquiring historical wind power data as an original input data set and preprocessing input data of the original input data set includes the following steps:
acquiring historical wind power data as an original input data set, and acquiring the maximum value of input data and the minimum value of the input data in the original input data set;
and normalizing the input data in the original input data set according to the maximum value of the input data and the minimum value of the input data.
Preferably, the step of performing wavelet time-frequency analysis on the processed input data, obtaining wavelet coefficients, and generating a corresponding two-dimensional time-frequency graph according to the time domain characteristics and the frequency domain characteristics includes the following steps:
acquiring wavelet coefficients under different frequencies according to the processed input data so as to extract time domain characteristics and frequency domain characteristics of historical time sequences;
and combining the wavelet coefficients under different frequencies according to time frequency to generate a wavelet time-frequency graph.
Preferably, the step of building a neural network prediction model for processing the time-frequency diagram according to the characteristics of the time-frequency diagram, the size of the original input data set, and the conditions of a hardware platform includes the following steps:
establishing an input layer according to the characteristics of the time-frequency diagram;
establishing an alternating structure of a plurality of convolution layers and pooling layers according to the input layer so as to extract image characteristics of the time-frequency graph;
establishing a full connection layer according to the convolution layer and the pooling layer, wherein the full connection layer reduces the dimension of a feature image obtained by processing the last pooling layer, converts the feature image into a feature vector and outputs a final prediction result;
and building a neural network prediction model according to the size of the original input data set, the condition of a hardware platform, the input layer, the full-connection layer, the plurality of alternately arranged convolution layers and the pooling layer.
Preferably, the step of performing, by the full-link layer, dimensionality reduction on the feature image obtained by processing the last pooling layer, and converting the feature image into a feature vector includes the steps of:
and finally, converting the characteristic image obtained by the processing of the pooling layer into a one-dimensional characteristic vector through a Flatten algorithm.
Preferably, the step of establishing an alternating structure of a plurality of convolutional layers and a pooling layer according to the input layer comprises the steps of:
according to the input layer, establishing a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a third convolution layer and a third pooling layer which are connected in sequence; the first convolution layer, the second convolution layer and the third convolution layer are used for eliminating invalid information of the time-frequency graph layer by layer and extracting key features; the first pooling layer, the second pooling layer and the third pooling layer are used for further extracting image features and compressing feature images;
respectively establishing convolution kernels for extracting image features for the first convolution layer, the second convolution layer and the third convolution layer, wherein each convolution kernel corresponds to one image feature;
determining a pooling method of the first, second, and third pooling layers.
Preferably, the activation function of the first convolution layer is a hyperbolic tangent activation function, and the activation function of the second convolution layer is a linear rectification activation function; the activation function of the third convolution layer is a hyperbolic tangent activation function.
Preferably, the convolution kernel size of the first convolution layer, the convolution kernel size of the second convolution layer, and the convolution kernel size of the third convolution layer are sequentially reduced.
Preferably, the first pooling layer, the second pooling layer and the third pooling layer use a maximal combination method to extract time-frequency features.
Preferably, the step of obtaining an image feature extraction result, a training error result, and a final prediction result according to the original input data set and the neural network prediction model, and adjusting parameters of the neural network prediction model according to the image feature extraction result, the training error result, and the final prediction result to obtain an optimal prediction result includes the following steps:
establishing a test set, a training set and a model input set according to the original input data set;
acquiring the image feature extraction result, the training error result and the final prediction result according to the test set, the training set and the neural network prediction model so as to adjust the learning rate and the training times of the neural network prediction model and the convolution kernel size of each convolution layer;
and obtaining a wind power prediction result according to the model input set and the adjusted neural network prediction model.
In the technical scheme of the invention, wavelet analysis is firstly adopted to convert a time sequence into a time-frequency graph sequence so as to fully utilize time domain characteristics and frequency domain characteristics of the time sequence, then a convolutional neural network is adopted to extract the time-frequency graph characteristics, a back propagation algorithm is utilized to optimize a network structure, and finally final output is obtained through regression. The method can fully utilize the time-frequency information of the time sequence on one hand and the excellent characteristic of two-dimensional convolution on the other hand, and realizes accurate ultra-short term prediction of the time sequence, thereby obtaining the accurate ultra-short term prediction result of the wind power.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.
FIG. 1 is a schematic diagram of a time series ultra-short term prediction model based on wavelet analysis and a convolutional neural network.
Fig. 2 is a schematic diagram of a convolutional neural network structure.
FIG. 3 is a graph of relationship between wind power and data points in No. 2/month and No. 12/2014
Fig. 4 is a time-frequency diagram of wind power No. 2/month No. 12 in 2014.
Fig. 5 is a graph of relationship between wind power and data points in No. 2/month and No. 13 in 2014.
Fig. 6 is a time-frequency diagram of the wind power of No. 2/month and No. 13 in 2014.
Fig. 7 is a graph of relationship between wind power and data points in No. 2/month and No. 14 in 2014.
Fig. 8 is a time-frequency diagram of wind power No. 2/month No. 14 in 2014.
FIG. 9 is a diagram illustrating convolution operations.
FIG. 10 is a flow chart of three different pooling layer sub-sampling training procedures.
Fig. 11 is a graph of visualization results of convolution layers of a Convolution Neural Network (CNN) model for performing ultra-short term prediction on wind power based on the ultra-short term wind power prediction method of wavelet time-frequency imaging.
Fig. 12 shows the corresponding recognition accuracy rates when different convolution kernels are selected for the first convolution layer, the second convolution layer and the third convolution layer for performing ultra-short-term prediction on wind power by the ultra-short-term wind power prediction method based on wavelet time-frequency imaging.
Fig. 13 is a prediction model training error curve when the model learning rate of the ultra-short term wind power prediction method based on wavelet time-frequency imaging is 0.01 for ultra-short term wind power prediction.
Fig. 14 is a prediction model training error curve when the model learning rate of the ultra-short term wind power prediction method based on wavelet time-frequency imaging is 0.012, for performing ultra-short term prediction on wind power.
Fig. 15 is a prediction model training error curve when the model learning rate of the ultra-short term wind power prediction method based on wavelet time-frequency imaging is 0.014.
Fig. 16 is a prediction model prediction result curve diagram of ultra-short term prediction of wind power by the ultra-short term wind power prediction method based on wavelet time-frequency imaging.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that all the directional indicators (such as up, down, left, right, front, and rear … …) in the embodiment of the present invention are only used to explain the relative position relationship between the components, the movement situation, etc. in a specific posture (as shown in the drawing), and if the specific posture is changed, the directional indicator is changed accordingly.
In addition, the descriptions related to "first", "second", etc. in the present invention are only for descriptive purposes and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the present invention, unless otherwise expressly stated or limited, the terms "connected," "secured," and the like are to be construed broadly, and for example, "secured" may be a fixed connection, a removable connection, or an integral part; can be mechanically or electrically connected; they may be directly connected or indirectly connected through intervening media, or they may be connected internally or in any other suitable relationship, unless expressly stated otherwise. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
In addition, the technical solutions in the embodiments of the present invention may be combined with each other, but it must be based on the realization of those skilled in the art, and when the technical solutions are contradictory or cannot be realized, such a combination of technical solutions should not be considered to exist, and is not within the protection scope of the present invention.
Referring to fig. 1 of fig. 1 to fig. 16, in order to achieve the above object, the method for predicting ultra-short-term wind power based on wavelet time-frequency imaging according to the present invention includes the following steps:
s100, acquiring historical wind power data as an original input data set, and preprocessing input data of the original input data set;
s200, performing wavelet time-frequency analysis on the processed input data, acquiring wavelet coefficients, and generating a corresponding two-dimensional time-frequency graph according to time domain characteristics and frequency domain characteristics;
s300, building a neural network prediction model for processing the time-frequency diagram according to the characteristics of the time-frequency diagram, the original input data set and the conditions of a hardware platform;
s400, obtaining an image feature extraction result, a training error result and a final prediction result according to an original input data set and the neural network prediction model, and adjusting parameters of the neural network prediction model according to the obtained image feature extraction result, the training error result and the final prediction result to obtain an optimal wind power ultra-short term prediction result. In the technical scheme of the invention, wavelet analysis is firstly adopted to convert a time sequence into a time-frequency graph, time domain characteristics and frequency domain characteristics of the time sequence are fully utilized, then a convolutional neural network is adopted to extract the time-frequency graph characteristics, a back propagation algorithm is utilized to optimize a network structure, and finally final output is obtained through regression. The method can fully utilize the time-frequency information of the time sequence on one hand and the excellent characteristic of two-dimensional convolution on the other hand, and realizes accurate ultra-short term prediction of the time sequence, thereby obtaining the accurate ultra-short term prediction result of the wind power.
Specifically, the method firstly performs normalization preprocessing on the original time sequence, facilitates subsequent data use and rapid convergence of a prediction program, simultaneously extracts time domain characteristics and frequency domain characteristics of the sequence from the preprocessed time sequence through wavelet time-frequency analysis, images the time domain characteristics and the frequency domain characteristics of the time sequence, and can accurately reflect local characteristics of the time sequence in a time-frequency domain.
Due to the weight sharing structure of the convolutional neural network, the convolutional neural network has high invariance to the translation, scaling, inclination and other forms of deformation of the image, the early-stage complex preprocessing process of the image is effectively avoided, the original image can be directly processed, the time-frequency graph obtained by wavelet time-frequency analysis through the convolutional neural network processing can fully utilize the excellent characteristics of the convolutional neural network, the time-frequency graph characteristics are extracted, the accurate prediction of a time sequence is completed, the problems that the statistic of noise is difficult to estimate and LS-SVM lacks sparsity in a common Kalman filtering method are solved, and the convolutional neural network has wide application prospect.
Referring to fig. 1, based on the first embodiment of the ultra-short term wind power prediction method based on wavelet time-frequency imaging of the present invention, and the second embodiment of the ultra-short term wind power prediction method based on wavelet time-frequency imaging of the present invention, the step S100 includes the following steps:
s110, acquiring historical wind power data as an original input data set, and acquiring the maximum value of input data and the minimum value of the input data in the original input data set;
and S120, normalizing the input data in the original input data set according to the maximum value of the input data and the minimum value of the input data.
Specifically, in order to avoid that the unit magnitude of the input data is too large and affects the subsequent prediction accuracy, normalization processing needs to be performed on the data in advance, which is a step for facilitating subsequent data processing and ensuring fast convergence of a program.
The input data and the output data after normalization processing are controlled within the range of [0,1], and the specific operation of normalization is shown as formula (1).
Figure RE-GDA0002787000040000071
In the formula: p is a radical ofnRaw input power data; p is a radical ofminRepresents the minimum in the original dataset; p is a radical ofmaxRepresents the maximum value in the original dataset;
Figure RE-GDA0002787000040000072
normalized power data.
Referring to fig. 3 to 8, in a second embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, and in a third embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, the step S200 includes the following steps:
s210, acquiring wavelet coefficients under different frequencies according to the processed input data to extract time domain characteristics and frequency domain characteristics of a historical time sequence;
and S220, combining the wavelet coefficients under different frequencies according to time frequency to generate a wavelet time-frequency graph.
Specifically, when a function ψ (t) ∈ L2(R)(L2(R) space where signal energy is limited) Fourier transform ψ (ω)*When the condition shown in the following formula (2) is satisfied, the function is a wavelet basis function, the basis function can obtain a wavelet sequence through translation and stretching, the continuous wavelet sequence is shown in formula (3), and the discrete wavelet sequence is shown in formula (4).
Figure RE-GDA0002787000040000081
Figure RE-GDA0002787000040000082
In the formula: a is a scale factor; b is a time shift factor.
ψj,t(t)＝2-j/2ψ(2-j-k)，j,k∈R；j≠0 (4)；
The wavelet basis functions with scale factors and time shift factors can focus on different parts of data by changing the length and the position of wavelets.
Thus, at low frequencies, wavelet analysis has a higher frequency resolution and a lower time resolution; in the high frequency part, the wavelet analysis has a higher time resolution and a lower frequency resolution.
The high-precision local feature extraction function is the most important and practical feature of the wavelet analysis method. The wavelet analysis method can select different wavelet basis functions aiming at different engineering objects.
In the wavelet operation process, a wavelet function starts from the starting point of a signal, convolution is carried out on window signals at different positions through continuous movement on a time axis, wavelet coefficients corresponding to local signals are obtained through comparison until the convolution operation is completed on the whole signal, then the selected wavelet function is subjected to scale expansion by one unit, and the process is repeated until all scales are completed. The magnitude of the wavelet coefficient indicates how similar the waveform of the wave function and the waveform of the local signal at that moment are. The larger the wavelet coefficient is, the higher the waveform similarity of the selected wavelet waveform and the local signal at the moment is indicated.
In the frequency domain, wavelets of different lengths and frequencies are obtained by stretching and compressing the length of the wavelet. In the high frequency band, the wavelet length is compressed, the wavelet coefficient is reduced, the window is narrowed, the time resolution is improved, and the frequency resolution is reduced; at low frequencies, the wavelet length is stretched, the window becomes wider, the temporal resolution is reduced, and the frequency resolution is increased.
The wavelet time-frequency diagram is obtained by combining wavelet coefficients at different frequencies. Different wavelet basis functions are selected for different time series. dbN the wavelet has outstanding ability to extract the features of the unstable signal, and can be widely used for the analysis of the unstable signal.
Referring to fig. 1 and fig. 2, based on the first embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, in a fourth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, the step S300 includes:
s310, establishing an input layer according to the characteristics of the time-frequency diagram;
s320, establishing an alternating structure of a plurality of convolutional layers and pooling layers according to the input layer to extract the image characteristics of the time-frequency graph;
s330, establishing a full connection layer according to the convolution layer and the pooling layer, wherein the full connection layer reduces the dimension of the feature image obtained by processing the last pooling layer, converts the feature image into a feature vector and outputs a final prediction result;
s340, building a neural network prediction model according to the size of the original input data set, the condition of a hardware platform, the input layer, the full-connection layer, the plurality of alternately arranged convolution layers and the pooling layer.
Specifically, as can be seen from fig. 2, the basic structure of the convolutional neural network includes several parts, i.e., an input layer, a convolutional layer, a pooling layer, a fully-connected layer, and an output layer, each of the convolutional neural networks has an alternating structure of a plurality of convolutional layers and pooling layers, and this structure is equivalent to an "image filter" and is used for extracting image features, and the conversion from low-dimensional features to high-dimensional features of an image is realized through continuous filtering by the filter. The full-connection layer is equivalent to a classifier, receives the image transmitted from the front end, reduces the dimension of the characteristic image transmitted from the front end through a corresponding algorithm, converts the characteristic image into a characteristic vector, and then outputs a final prediction result. The convolutional layer is a key part of a convolutional neural network different from a traditional neural network and is mainly used for feature extraction of images. When image processing is performed, it is necessary to perform a convolution operation on an image by using a filter (also referred to as a convolution kernel) to extract image features.
As shown in fig. 9, the specific operation is to multiply the matrix formed by the adjacent pixels of the image by the corresponding elements of the matrix of the filter, add the obtained products to obtain the processed pixel values, and sequentially slide the filter until all the pixels complete the convolution operation. For the image edge, there is no way to directly perform convolution operation, edge elements need to be filled, and the filled pixel value is generally selected to be 0 or the central pixel value, so as to ensure the normal operation of convolution operation.
Referring to fig. 1, in a fourth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, and in a fifth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, the step of performing, by the full link layer in step S330, dimension reduction on the feature image obtained by processing the last pooling layer, and converting the feature image into a feature vector includes the following steps:
and finally, converting the characteristic image obtained by the processing of the pooling layer into a one-dimensional characteristic vector through a Flatten algorithm.
Specifically, the last pooling layer converts the processed two-dimensional image into a one-dimensional feature vector through a Flatten algorithm, and finally, the ultra-short-term prediction of a time sequence is realized through a full-connection layer.
Referring to fig. 1, in a sixth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, the step of establishing an alternating structure of a plurality of convolution layers and pooling layers according to the input layer in step S320 includes:
s321, establishing a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a third convolution layer and a third pooling layer which are connected in sequence according to the input layer; the first convolution layer, the second convolution layer and the third convolution layer are used for eliminating invalid information of the time-frequency graph layer by layer and extracting key features, and the first pooling layer, the second pooling layer and the third pooling layer are used for further extracting image features and compressing feature images;
s322, respectively establishing convolution kernels for extracting image features for the first convolution layer, the second convolution layer and the third convolution layer, wherein each convolution kernel corresponds to one image feature;
s323, determining the pooling methods of the first pooling layer, the second pooling layer and the third pooling layer.
Specifically, the model increases the network capacity, reduces the feature resolution and the number of network parameters to be optimized through the alternating structure of the convolutional layer and the pooling layer, and realizes feature extraction on the time-frequency feature map. As shown in fig. 1, a 7-layer convolutional neural network model (including sequentially connecting a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, and a fully-connected layer) is constructed, as shown in fig. 1.
Referring to fig. 1, in a sixth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, and in a seventh embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, the activation function of the first convolution layer is a hyperbolic tangent activation function (tanh) activation function, and the activation function of the second convolution layer is a linear rectification function (ReLU) activation function; the activation function of the third convolution layer is a hyperbolic tangent activation function (tanh) activation function.
Specifically, the neural network generally deals with the nonlinear problem, and the activation function is to compensate for the expression capability of the linear model and add a certain nonlinear factor to the neural network.
Common activation functions are sigmoid function, hyperbolic tangent (tanh) function and linear rectification (ReLU) function.
The sigmoid function is shown as follows:
Figure RE-GDA0002787000040000111
the output interval is (0, 1), and the output result is monotonously continuous without taking 0 as the output center, but the function itself has soft saturation, so that the problem of gradient disappearance is easy to appear, and the method is generally used for an output layer.
The hyperbolic tangent function is shown below:
Figure RE-GDA0002787000040000112
the output interval is (-1,1), 0 is taken as an output center, the hyperbolic tangent function and the sigmoid function have the problem of gradient disappearance, but the training speed is improved to a certain extent, and the method is generally used for hidden layers.
The linear rectification ReLU function is shown as follows:
Figure RE-GDA0002787000040000113
when the input is less than 0, the output is 0, and when the input is more than 0, the data is output, so that the problem of gradient explosion of the first two activation functions is solved, the training speed is improved, the performance is good in unsupervised learning, and the method is an activation function widely applied at the present stage.
Referring to fig. 1 and 9, in an eighth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging, a convolution kernel size of the first convolution layer, a convolution kernel size of the second convolution layer, and a convolution kernel size of the third convolution layer are sequentially reduced.
Specifically, the training effects are different due to different sizes of convolution kernels, and the smaller the scale of the convolution kernels is, the faster the corresponding model training speed is, the higher the model training loss is, and the lower the feature recognition accuracy rate is; the larger the scale of the convolution kernel is, the more gradual the corresponding model training speed is, the lower the model training loss is, and the feature recognition accuracy (auray) is relatively improved. And selecting the convolution kernel scale of each convolution layer according to the feature identification accuracy rate of the convolution kernel with different scales corresponding to different convolution layers.
Referring to fig. 1 and fig. 2, in a sixth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, in a ninth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, the first pooling layer, the second pooling layer, and the third pooling layer all use a maximal combination method to extract time-frequency features.
Specifically, the pooling layer after the convolution layer is a structure for further extracting image features, compressing an image, reducing the amount of calculation, and preventing overfitting of a model. The pooling layer achieves the goal of scaling down the data size mainly by sub-sampling.
For example, sub-sampling a 4 × 4 image by 2 × 2 will obtain a 2 × 2 image, which is equivalent to extracting a representative pixel from every 4 pixels in the original image as an output, and greatly reduces the amount of calculation of the whole model.
Common pooling methods are max-combining, mean-combining, and random-combining, and the max-combining method is generally used more frequently, as shown in fig. 10.
Referring to fig. 1, based on the above-mentioned embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, in a tenth embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to the present invention, the step S400 includes the following steps:
s410, establishing a test set, a training set and a model input set according to the original input data set;
s420, obtaining the image feature extraction result, the training error result and the final prediction result according to the test set, the training set and the neural network prediction model, and adjusting the learning rate and the training times of the neural network prediction model and the convolution kernel size of each convolution layer;
and S430, obtaining an optimal wind power ultra-short term prediction result according to the model input set and the adjusted neural network prediction model. Specifically, the whole training process of the prediction model adopts a batch gradient descent algorithm, so that the network structure can be optimized by using a reverse regression algorithm, and the BP algorithm (namely a back propagation algorithm) is suitable for a learning algorithm of a multilayer neuron network and is established on the basis of a gradient descent method. The input-output relationship of the BP network is substantially a mapping relationship: an n-input m-output BP neural network performs the function of continuous mapping from n-dimensional euclidean space to a finite field in m-dimensional euclidean space, which is highly non-linear. Its information processing ability comes from multiple composition of simple non-linear function, so it has strong function reproduction ability.
In another embodiment of the ultra-short-term wind power prediction method based on wavelet time-frequency imaging, the actual calculation example of ultra-short-term prediction of wind power is performed by using the ultra-short-term wind power prediction method based on wavelet time-frequency imaging.
The data set used in the method is winter data from No. 1 month 1 to No. 3 month 1 month 2014 of a certain wind power plant in northeast China, in the time period, the climate of the wind power plant does not change obviously, and the wind power generation data are less influenced by weather. The wind power station collects wind power values at intervals of 15 minutes, 5000 data points are selected as an original input data set in total, normalization processing is carried out on the original input data set, and the obtained data are used for subsequent prediction.
The data set for month 1 is referred to herein as the training set and the data for month 2 is referred to herein as the test set. The input set of the model is a three-day wind power historical time sequence, the image size of the obtained time-frequency graph is uniformly adjusted to 300 x 150 for facilitating subsequent processing, and the output is the generated power of a wind power plant in the future 15 minutes.
dbN the wavelet has outstanding ability to extract the features of the unstable signal, and can be widely used for the analysis of the unstable signal. Because the wind power historical time sequence belongs to unstable signals, the time-frequency characteristics of the wind power historical time sequence under different frequencies are extracted by adopting tightly-supported orthogonal wavelets (db6 wavelets), and the db6 wavelet has the excellent characteristics of good regularity, large moment of disappearance and the like. And performing time-frequency analysis on the wind power of different frequency bands by telescopic translation of the db6 wavelet, and extracting local frequency domain characteristics of the data.
First, feature extraction analysis
Selecting a wind power historical time sequence of No. 12 to No. 14 three days in No. 2 month in 2014, and performing wavelet time-frequency joint analysis on the wind power historical time sequence of 96 data points (24 multiplied by 60 divided by 15) each day, wherein a time-frequency analysis chart is shown in the following fig. 3-8.
As shown in fig. 1 and 11, the first convolution layer, the second convolution layer, and the third convolution layer respectively establish 4 convolution kernels, 8 convolution kernels, and 16 convolution kernels, and one convolution kernel corresponds to one image feature.
Therefore, 4, 8 and 16 image features are extracted from the first convolutional layer, the second convolutional layer and the third convolutional layer, respectively, and for further understanding the convolutional neural network feature extraction process, the image features extracted from each layer are visualized by a code reversal method, the image features are visualized by the first convolutional layer, the second convolutional layer and the third convolutional layer, respectively, and the features extracted by different convolutional kernels are also different, as shown in fig. 6.
The edge features of the image are extracted from the first convolution layer, basic information of the image is reserved, the approximate outline of the image can be seen, feature _ maps corresponding to different convolution kernels are different to a certain extent, and different convolution kernels are proved to identify different image features.
With the continuous deepening of the layer number, the image features extracted by the convolution layer are more and more abstract, invalid information is further filtered, and the blank content is more because the second convolution kernel and the third convolution kernel do not find matched image texture features. And eliminating invalid information layer by layer from the time-frequency image through the first convolution layer, the second convolution layer and the third convolution layer, and extracting key features.
The recognition accuracy rates corresponding to different convolution kernels selected for the first convolution layer, the second convolution layer and the third convolution layer are shown in fig. 12.
The accuracy for selecting different convolution kernels for each convolution layer is shown in table 1.
TABLE 1 feature recognition accuracy for convolutional layer selection for different convolutional kernels
Figure RE-GDA0002787000040000141
As can be seen from fig. 7 and table 1, the feature recognition accuracy of the first convolution layer is 94.912% with a convolution kernel size of 8 × 8; when the convolution kernel size of the second convolution layer is 4 multiplied by 4, the feature recognition accuracy rate is the highest and is 91.791%; the third convolutional layer has the highest feature recognition accuracy of 92.413% when the convolutional kernel size is 2 × 2. Therefore, the first convolution layer, the second convolution layer and the third convolution layer sequentially adopt convolution kernels with the sizes of 8 × 8, 4 × 4 and 2 × 2.
Second, training error analysis
The predictive model training error curves are shown in fig. 13-15.
As is clear from fig. 14, when the learning rate is 0.012, the error between the model training set and the test set is minimum, and the error in the test set is lower than that in the training set, so the learning rate of the model is 0.012.
The whole training process of the prediction model adopts a batch gradient descent algorithm, and the error loss of the model is given by an absolute average error.
The model training times are important factors influencing the prediction accuracy, and excessive training times can cause overfitting of training data; too low a number of training cycles may result in poor fit of the training data.
As can be seen from fig. 14, the training error and the testing error of the model always keep a stable descending trend, and the model error stabilizes around 0.01 at about 80 times. Therefore, the number of training times of the model is selected to be 80.
Analysis of predicted results
The prediction results of the prediction model are shown in fig. 16.
It can also be seen from fig. 16 that the fitting effect of the prediction curve output based on the model and the actual wind power curve is basically expected, although some places have partial entrances and exits, the general trend is basically consistent, the prediction can be achieved, and the model prediction effect is better.
The root mean square error of the model (square root of the ratio of the square of the deviation of the predicted value from the true value to the number of observations) was 0.241.
The method proves that the accuracy of the super-short-term prediction of the wind power time series can be effectively improved in a mode of combining wavelet time-frequency joint analysis and a convolutional neural network.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a computer-readable storage medium (such as ROM/RAM, magnetic disk, optical disk) as described above, and includes several instructions for enabling a terminal device to enter the method according to the embodiments of the present invention.
In the description herein, references to the description of the term "one embodiment," "another embodiment," or "first through xth embodiments," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, method steps, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
The above description is only for the preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all equivalent structures or equivalent flow transformations that may be embodied in the present specification and drawings, or may be directly or indirectly applied to other related arts.

Claims (10)

1. An ultrashort-term wind power prediction method based on wavelet time-frequency imaging is characterized by comprising the following steps:
acquiring historical wind power data as an original input data set, and preprocessing input data of the original input data set;
performing wavelet time-frequency analysis on the processed input data, acquiring wavelet coefficients, and generating a corresponding two-dimensional time-frequency graph according to time domain characteristics and frequency domain characteristics;
building a neural network prediction model for processing the time-frequency diagram according to the characteristics of the time-frequency diagram, the size of the original input data set and the conditions of a hardware platform;
obtaining an image feature extraction result, a training error result and a final prediction result according to an original input data set and the neural network prediction model, and adjusting parameters of the neural network prediction model according to the image feature extraction result, the training error result and the final prediction result to obtain an optimal wind power ultra-short term prediction result.
2. The ultra-short-term wind power prediction method based on wavelet time-frequency imaging as claimed in claim 1, wherein said step of obtaining historical wind power data as an original input data set and preprocessing input data of said original input data set comprises the steps of:
acquiring historical wind power data as an original input data set, and acquiring the maximum value of input data and the minimum value of the input data in the original input data set;
and normalizing the input data in the original input data set according to the maximum value of the input data and the minimum value of the input data.
3. The ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to claim 1, wherein the step of performing wavelet time-frequency analysis on the processed input data and obtaining wavelet coefficients, and generating a corresponding two-dimensional time-frequency map according to time-domain features and frequency-domain features comprises the steps of:
acquiring wavelet coefficients under different frequencies according to the processed input data so as to extract time domain characteristics and frequency domain characteristics of historical time sequences;
and combining the wavelet coefficients under different frequencies according to time frequency to generate a wavelet time-frequency graph.
4. The ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to claim 1, wherein the step of building a neural network prediction model for processing the time-frequency image according to the characteristics of the time-frequency image, the size of the original input data set and the conditions of a hardware platform comprises the following steps:
establishing an input layer according to the characteristics of the time-frequency diagram;
establishing an alternating structure of a plurality of convolution layers and pooling layers according to the input layer so as to extract image characteristics of the time-frequency graph;
establishing a full connection layer according to the convolution layer and the pooling layer, wherein the full connection layer reduces the dimension of a feature image obtained by processing the last pooling layer, converts the feature image into a feature vector and outputs a final prediction result;
and building a neural network prediction model according to the size of the original input data set, the condition of a hardware platform, the input layer, the full-connection layer, the plurality of alternately arranged convolution layers and the pooling layer.
5. The ultra-short-term wind power prediction method based on wavelet time-frequency imaging as claimed in claim 4, wherein said full link layer performs dimension reduction on the feature image processed by the last pooling layer, and the step of converting said feature image into a feature vector comprises the steps of:
and finally, converting the characteristic image obtained by the processing of the pooling layer into a one-dimensional characteristic vector through a Flatten algorithm.
6. The ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to claim 4, wherein the step of establishing an alternating structure of a plurality of convolutional layers and pooling layers according to the input layer comprises the steps of:
according to the input layer, establishing a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a third convolution layer and a third pooling layer which are connected in sequence; the first convolution layer, the second convolution layer and the third convolution layer are used for eliminating invalid information of the time-frequency graph layer by layer and extracting key features; the first pooling layer, the second pooling layer and the third pooling layer are used for further extracting image features and compressing feature images;
respectively establishing convolution kernels for extracting image features for the first convolution layer, the second convolution layer and the third convolution layer, wherein each convolution kernel corresponds to one image feature;
determining a pooling method of the first, second, and third pooling layers.
7. The wavelet time-frequency imaging-based ultrashort-term wind power prediction method of claim 6, wherein the activation function of the first convolution layer is a hyperbolic tangent activation function, and the activation function of the second convolution layer is a linear rectification activation function; the activation function of the third convolution layer is a hyperbolic tangent activation function.
8. The wavelet time-frequency imaging-based ultrashort-term wind power prediction method of claim 6, wherein the convolution kernel size of the first convolution layer, the convolution kernel size of the second convolution layer and the convolution kernel size of the third convolution layer are sequentially reduced.
9. The wavelet time-frequency imaging-based ultrashort-term wind power prediction method of claim 6, wherein the first pooling layer, the second pooling layer and the third pooling layer all use a maximal combination method to extract time-frequency features.
10. The ultra-short-term wind power prediction method based on wavelet time-frequency imaging according to any one of claims 1-9, wherein the step of obtaining an image feature extraction result, a training error result, and a final prediction result according to an original input data set and the neural network prediction model, and adjusting parameters of the neural network prediction model according to the image feature extraction result, the training error result, and the final prediction result to obtain an optimal prediction result comprises the steps of:
establishing a test set, a training set and a model input set according to the original input data set;
acquiring the image feature extraction result, the training error result and the final prediction result according to the test set, the training set and the neural network prediction model so as to adjust the learning rate and the training times of the neural network prediction model and the convolution kernel size of each convolution layer;
and obtaining an optimal wind power ultra-short term prediction result according to the model input set and the adjusted neural network prediction model.
CN202011232819.2A 2020-11-06 2020-11-06 Ultra-short-term wind power prediction method based on wavelet time-frequency imaging Active CN112348255B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011232819.2A CN112348255B (en) 2020-11-06 2020-11-06 Ultra-short-term wind power prediction method based on wavelet time-frequency imaging

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011232819.2A CN112348255B (en) 2020-11-06 2020-11-06 Ultra-short-term wind power prediction method based on wavelet time-frequency imaging

Publications (2)

Publication Number Publication Date
CN112348255A true CN112348255A (en) 2021-02-09
CN112348255B CN112348255B (en) 2024-04-09

Family

ID=74429739

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011232819.2A Active CN112348255B (en) 2020-11-06 2020-11-06 Ultra-short-term wind power prediction method based on wavelet time-frequency imaging

Country Status (1)

Country Link
CN (1) CN112348255B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115081489A (en) * 2022-07-13 2022-09-20 重庆大学 Time sequence classification method based on wavelet decomposition matrix and residual error network
CN117313927A (en) * 2023-09-19 2023-12-29 华能澜沧江水电股份有限公司 Wind power generation power prediction method and system based on wavelet neural network

Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1288766A2 (en) * 2001-08-31 2003-03-05 Sony United Kingdom Limited Digital content distribution
CN102102626A (en) * 2011-01-30 2011-06-22 华北电力大学 Method for forecasting short-term power in wind power station
JP2013114625A (en) * 2011-11-30 2013-06-10 A T Communications Co Ltd Mobile communication terminal, settlement system, settlement method, and program
CN105184391A (en) * 2015-08-19 2015-12-23 国网山东省电力公司电力科学研究院 Method for predicting wind speed and power of wind farm based on wavelet decomposition and support vector machine
CN106384170A (en) * 2016-09-24 2017-02-08 华北电力大学（保定） Wavelet decomposition and reconstruction-based time sequence wind speed prediction method
CN107292446A (en) * 2017-07-03 2017-10-24 西南交通大学 A kind of mixing wind speed forecasting method based on consideration component relevance wavelet decomposition
CN107507097A (en) * 2017-07-03 2017-12-22 上海电力学院 A kind of short-term wind power prediction method
WO2018028255A1 (en) * 2016-08-11 2018-02-15 深圳市未来媒体技术研究院 Image saliency detection method based on adversarial network
CN108448610A (en) * 2018-03-12 2018-08-24 华南理工大学 A kind of short-term wind power prediction method based on deep learning
CN109214575A (en) * 2018-09-12 2019-01-15 河海大学 A kind of super short-period wind power prediction technique based on small wavelength short-term memory network
CN109272156A (en) * 2018-09-12 2019-01-25 河海大学 A kind of super short-period wind power probability forecasting method
CN110045348A (en) * 2019-05-05 2019-07-23 应急管理部上海消防研究所 A kind of human motion state classification method based on improvement convolutional neural networks
CN110082841A (en) * 2019-04-18 2019-08-02 东华大学 A kind of short-term wind speed forecasting method
CN110415180A (en) * 2019-06-10 2019-11-05 西安电子科技大学 A kind of SAR image denoising method based on wavelet convolution neural network
CN110618353A (en) * 2019-10-22 2019-12-27 北京航空航天大学 Direct current arc fault detection method based on wavelet transformation + CNN
CN111091233A (en) * 2019-11-26 2020-05-01 江苏科技大学 Wind power plant short-term wind power prediction modeling method based on wavelet analysis and multi-model AdaBoost depth network
WO2020156348A1 (en) * 2019-01-31 2020-08-06 青岛理工大学 Structural damage identification method based on ensemble empirical mode decomposition and convolution neural network
CN111507884A (en) * 2020-04-19 2020-08-07 衡阳师范学院 Self-adaptive image steganalysis method and system based on deep convolutional neural network

Patent Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1288766A2 (en) * 2001-08-31 2003-03-05 Sony United Kingdom Limited Digital content distribution
CN102102626A (en) * 2011-01-30 2011-06-22 华北电力大学 Method for forecasting short-term power in wind power station
JP2013114625A (en) * 2011-11-30 2013-06-10 A T Communications Co Ltd Mobile communication terminal, settlement system, settlement method, and program
CN105184391A (en) * 2015-08-19 2015-12-23 国网山东省电力公司电力科学研究院 Method for predicting wind speed and power of wind farm based on wavelet decomposition and support vector machine
WO2018028255A1 (en) * 2016-08-11 2018-02-15 深圳市未来媒体技术研究院 Image saliency detection method based on adversarial network
CN106384170A (en) * 2016-09-24 2017-02-08 华北电力大学（保定） Wavelet decomposition and reconstruction-based time sequence wind speed prediction method
CN107292446A (en) * 2017-07-03 2017-10-24 西南交通大学 A kind of mixing wind speed forecasting method based on consideration component relevance wavelet decomposition
CN107507097A (en) * 2017-07-03 2017-12-22 上海电力学院 A kind of short-term wind power prediction method
CN108448610A (en) * 2018-03-12 2018-08-24 华南理工大学 A kind of short-term wind power prediction method based on deep learning
CN109214575A (en) * 2018-09-12 2019-01-15 河海大学 A kind of super short-period wind power prediction technique based on small wavelength short-term memory network
CN109272156A (en) * 2018-09-12 2019-01-25 河海大学 A kind of super short-period wind power probability forecasting method
WO2020156348A1 (en) * 2019-01-31 2020-08-06 青岛理工大学 Structural damage identification method based on ensemble empirical mode decomposition and convolution neural network
CN110082841A (en) * 2019-04-18 2019-08-02 东华大学 A kind of short-term wind speed forecasting method
CN110045348A (en) * 2019-05-05 2019-07-23 应急管理部上海消防研究所 A kind of human motion state classification method based on improvement convolutional neural networks
CN110415180A (en) * 2019-06-10 2019-11-05 西安电子科技大学 A kind of SAR image denoising method based on wavelet convolution neural network
CN110618353A (en) * 2019-10-22 2019-12-27 北京航空航天大学 Direct current arc fault detection method based on wavelet transformation + CNN
CN111091233A (en) * 2019-11-26 2020-05-01 江苏科技大学 Wind power plant short-term wind power prediction modeling method based on wavelet analysis and multi-model AdaBoost depth network
CN111507884A (en) * 2020-04-19 2020-08-07 衡阳师范学院 Self-adaptive image steganalysis method and system based on deep convolutional neural network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
刘欢;张超;: "基于小波-神经网络的短期风功率预测", 电子世界, no. 14 *
杨杰;霍志红;何永生;郭苏;邱良;许昌;: "基于小波与最小资源分配网络的超短期风电功率预测研究", 电力系统保护与控制, no. 09 *
林涛;赵参参;赵成林;刘航鹏;: "计及频率分析的风电场短期功率预测", 计算机仿真, no. 06 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115081489A (en) * 2022-07-13 2022-09-20 重庆大学 Time sequence classification method based on wavelet decomposition matrix and residual error network
CN117313927A (en) * 2023-09-19 2023-12-29 华能澜沧江水电股份有限公司 Wind power generation power prediction method and system based on wavelet neural network

Also Published As

Publication number Publication date
CN112348255B (en) 2024-04-09

Similar Documents

Publication Publication Date Title
NL2022758B1 (en) Image Super-resolution Reconstruction Method Based on Multi-scale Generative Adversarial Network
CN111012336B (en) Parallel convolutional network motor imagery electroencephalogram classification method based on spatio-temporal feature fusion
CN107316013B (en) Hyperspectral image classification method based on NSCT (non-subsampled Contourlet transform) and DCNN (data-to-neural network)
CN110599409B (en) Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel
CN110232341B (en) Semi-supervised learning image identification method based on convolution-stacking noise reduction coding network
CN103049892B (en) Non-local image denoising method based on similar block matrix rank minimization
CN107832787B (en) Radar radiation source identification method based on bispectrum self-coding characteristics
CN107633486A (en) Structure Magnetic Resonance Image Denoising based on three-dimensional full convolutional neural networks
CN112348255B (en) Ultra-short-term wind power prediction method based on wavelet time-frequency imaging
CN110287777B (en) Golden monkey body segmentation algorithm in natural scene
CN110533683B (en) Image omics analysis method fusing traditional features and depth features
CN106097278B (en) Sparse model, reconstruction method and dictionary training method of multi-dimensional signal
CN109190511B (en) Hyperspectral classification method based on local and structural constraint low-rank representation
CN108257093B (en) Single-frame image super-resolution method based on controllable kernel and Gaussian process regression
CN112818764B (en) Low-resolution image facial expression recognition method based on feature reconstruction model
CN116310459B (en) Hyperspectral image subspace clustering method based on multi-view spatial spectrum combination
CN111160392A (en) Hyperspectral classification method based on wavelet width learning system
CN114239657A (en) Time sequence signal identification method based on complex value interference neural network
CN116524253A (en) Thyroid cancer pathological image classification method based on lightweight transducer
CN104636711A (en) Facial emotion recognition method based on local sparse representation classifier
CN106296583B (en) Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps
CN117392153B (en) Pancreas segmentation method based on local compensation and multi-scale adaptive deformation
Xiong et al. Gradient boosting for single image super-resolution
Hel-Or et al. The role of redundant bases and shrinkage functions in image denoising
CN112733589B (en) Infrared image pedestrian detection method based on deep learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant