CN115564092A - Short-time wind power prediction system and method for wind power plant - Google Patents
Short-time wind power prediction system and method for wind power plant Download PDFInfo
- Publication number
- CN115564092A CN115564092A CN202211028966.7A CN202211028966A CN115564092A CN 115564092 A CN115564092 A CN 115564092A CN 202211028966 A CN202211028966 A CN 202211028966A CN 115564092 A CN115564092 A CN 115564092A
- Authority
- CN
- China
- Prior art keywords
- feature
- vector
- power
- data
- predicted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 239000013598 vector Substances 0.000 claims description 414
- 239000011159 matrix material Substances 0.000 claims description 191
- 238000000605 extraction Methods 0.000 claims description 54
- 230000004927 fusion Effects 0.000 claims description 34
- 238000013527 convolutional neural network Methods 0.000 claims description 25
- 238000003062 neural network model Methods 0.000 claims description 22
- 238000010586 diagram Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 8
- 238000012937 correction Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 3
- 238000005065 mining Methods 0.000 description 15
- 238000011161 development Methods 0.000 description 10
- 230000018109 developmental process Effects 0.000 description 10
- 230000006870 function Effects 0.000 description 9
- 238000010606 normalization Methods 0.000 description 8
- 230000008859 change Effects 0.000 description 7
- 230000003044 adaptive effect Effects 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000007405 data analysis Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000007704 transition Effects 0.000 description 3
- 238000010521 absorption reaction Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000013179 statistical model Methods 0.000 description 2
- 102100033620 Calponin-1 Human genes 0.000 description 1
- 102100033591 Calponin-2 Human genes 0.000 description 1
- 101000945318 Homo sapiens Calponin-1 Proteins 0.000 description 1
- 101000945403 Homo sapiens Calponin-2 Proteins 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Algebra (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Operations Research (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Computing Systems (AREA)
- Water Supply & Treatment (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to the field of wind power intelligent prediction of a wind power plant, and particularly discloses a short-time wind power prediction system and a short-time wind power prediction method for the wind power plant.
Description
Technical Field
The invention relates to the field of wind power intelligent prediction of a wind power plant, in particular to a short-time wind power prediction system and a short-time wind power prediction method for the wind power plant.
Background
Wind power is the most valuable renewable energy source and is widely developed and utilized worldwide. However, due to the randomness of the wind speed, the wind power has great fluctuation and uncertainty, which affects the stable operation of the power system, thereby limiting the absorption of the power system to the wind power.
According to the latest statistical report of the State energy agency, the method comprises the following steps: the installed capacity of the inner Mongolia wind power accounts for about 30% of the total amount of all energy sources in the area, but the capacity share of the wind power generation set on the power grid is less than 2% of the total installed capacity of the wind power. To improve the development and utilization of wind power, the accuracy of wind power prediction needs to be further improved.
Some existing wind power prediction schemes for wind power plants, for example, utilize historical meteorological information and historical power information to perform short-time power prediction, but the accuracy of power prediction is not high due to the fact that implicit mode features in the historical meteorological information, implicit mode features in the historical power information and associations between the historical meteorological information and the historical power information cannot be sufficiently mined by the data statistical model adopted by the existing wind power prediction schemes.
Therefore, an optimized short-term wind power prediction scheme for wind farms is desired.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a short-time wind power prediction system and a short-time wind power prediction method for a wind power plant, which adopt an artificial intelligence prediction technology, and take a convolutional neural network model as a feature extractor to perform deep mining on implicit dynamic change feature information of historical meteorological data and historical power data in a time dimension and correlation features between the historical meteorological information and the historical power information, so as to predict the power value of the wind power plant to be predicted at the next preset time point based on a feature retrieval mode of the historical data, and improve the development and utilization of wind power.
According to one aspect of the present application, there is provided a short-time wind power prediction system for a wind farm, comprising:
the historical data acquisition module is used for acquiring historical meteorological data and historical power data, wherein the meteorological data of each preset time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure;
the single-point meteorological data encoding module is used for arranging various data in the meteorological data of each preset time point in the historical meteorological data into input vectors and then obtaining single-point multi-dimensional meteorological feature vectors through a full connection layer;
the single-weather image data coding module is used for arranging the single-point multi-dimensional weather feature vectors into one-dimensional feature vectors by taking a single day as a time dimension and then obtaining multi-scale single-weather image feature vectors corresponding to each day through the multi-scale neighborhood feature extraction module;
the global meteorological data coding module is used for arranging the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix and then obtaining a historical meteorological feature matrix through a first convolution neural network model serving as a feature extractor;
the single-day power data coding module is used for arranging the power data of all the preset time points in each day of the historical power data into power input vectors and then obtaining single-day multi-scale power characteristic vectors corresponding to each day through the multi-scale neighborhood characteristic extraction module;
the global power data coding module is used for arranging the single-day multi-scale power characteristic vectors corresponding to each day into a two-dimensional characteristic matrix and then obtaining a historical power characteristic matrix through a second convolutional neural network serving as a characteristic extractor;
the historical data feature fusion module is used for fusing the historical meteorological feature matrix and the historical power feature matrix to obtain a fusion feature matrix;
the system comprises a data acquisition module on the day to be predicted, a data acquisition module on the day to be predicted and a data acquisition module on the day to be predicted, wherein the data acquisition module is used for acquiring meteorological data of a plurality of preset time points on the day to be predicted and output power of a wind power plant to be predicted at the preset time points;
the to-be-predicted data encoding module is used for converting meteorological data of a plurality of preset time points on the day to be predicted into to-be-predicted meteorological feature vectors, converting output power of the wind power plant to be predicted at the preset time points into to-be-predicted power feature vectors, and fusing the to-be-predicted meteorological feature vectors and the to-be-predicted power feature vectors to obtain predicted feature vectors;
the feature query module is used for multiplying the predicted feature vector and the fusion feature matrix to obtain a decoding feature vector;
the correction module is used for correcting the characteristic values of all positions in the decoding characteristic vector based on the mean value and the variance of the characteristic value sets of all the positions of the decoding characteristic vector to obtain a corrected decoding characteristic vector;
and the predicted value generation module is used for performing decoding regression on the corrected decoded characteristic vector through a decoder to obtain a decoded value, and the decoded value is used for representing the power predicted value of the wind power plant to be predicted at the next preset time point.
In the above short-term wind power generation system for wind power plantIn the measurement system, the single-point meteorological data encoding module is further configured to: arranging various data in the meteorological data of each preset time point in the historical meteorological data into an input vector according to a time dimension; and performing full-connection coding on the input vector by using the full-connection layer according to the following formula to extract high-dimensional implicit features of feature values of all positions in the input vector so as to obtain the single-point multi-dimensional meteorological feature vector, wherein the formula is as follows:wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the offset vector,representing a matrix multiplication.
In the above short-term wind power prediction system for a wind farm, the single weather image data encoding module includes: the first convolution unit is used for inputting the one-dimensional feature vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale single weather image feature vector, wherein the first convolution layer is provided with a first one-dimensional convolution kernel with a first length; a second convolution unit, configured to input the one-dimensional feature vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale single weather image feature vector, where the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and the cascading unit is used for cascading the first neighborhood scale single weather image feature vector and the second neighborhood scale single weather image feature vector to obtain the multi-scale single weather image feature vector corresponding to each day.
In the above short-time wind power prediction system for a wind farm, the first convolution unit is further configured to; performing one-dimensional convolution coding on the one-dimensional feature vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first neighborhood scale single weather image feature vector;
wherein the formula is:
wherein a is the width of the first convolution kernel in the X direction, F is a first convolution kernel parameter vector, G is a local vector matrix operated with a convolution kernel function, w is the size of the first convolution kernel, and X represents a tension input vector; the second convolution unit further configured to: performing one-dimensional convolution coding on the one-dimensional feature vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second neighborhood scale single weather image feature vector;
wherein the formula is:
wherein b is the width of the second convolution kernel in the X direction, F is a parameter vector of the second convolution kernel, G is a local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents a tension input vector.
In the above short-time wind power prediction system for a wind farm, the global meteorological data encoding module is further configured to: using each layer of the first convolutional neural network model as a feature extractor to respectively perform the following steps on input data in the forward direction transfer of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on local channel dimensions on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the first convolution neural network model as the feature extractor is the historical meteorological feature matrix, and the input of the first layer of the first convolution neural network model as the feature extractor is the two-dimensional feature matrix.
In the above short-time wind power prediction system for a wind farm, the historical data feature fusion module is further configured to: fusing the historical meteorological feature matrix and the historical power feature matrix according to the following formula to obtain a fused feature matrix;
wherein the formula is:
M 1 =M*M 2
wherein M is 1 Representing said historical meteorological features matrix, M 2 Representing the historical power feature matrix, and M representing the fusion feature matrix.
In the above short-time wind power prediction system for a wind farm, the correction module is further configured to: based on the mean and variance of the feature value sets of all positions of the decoding feature vector, correcting the feature value of each position in the decoding feature vector by the following formula to obtain the corrected decoding feature vector;
wherein the formula is:
wherein v is i Representing the eigenvalues of the respective positions in the decoded eigenvector, V representing the decoded eigenvector, mu and sigma being the sets of features V, respectively i E.g., the mean and variance of V, and L is the length of the decoded feature vector, and α is a weight hyperparameter.
In the above short-time wind power prediction system for a wind farm, the prediction value generation module is further configured to: decoding the corrected decoded feature vector using a plurality of fully connected layers of the decoder to perform a decoding regression on the corrected decoded feature vector to obtain the decoded value, wherein the formula is: wherein X is the corrected decoded feature vector, Y is the decoded value, W is a weight momentThe matrix, B, is the offset vector,representing the matrix multiplication, h (-) is the activation function.
According to another aspect of the application, a short-time wind power prediction method for a wind farm includes:
acquiring historical meteorological data and historical power data, wherein the meteorological data of each preset time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure;
arranging all data in the meteorological data of all preset time points in the historical meteorological data into input vectors, and then obtaining single-point multi-dimensional meteorological feature vectors through a full connection layer;
arranging the single-point multi-dimensional meteorological feature vectors into one-dimensional feature vectors by taking a single day as a time dimension, and then obtaining multi-scale single-meteorological feature vectors corresponding to each day through a multi-scale neighborhood feature extraction module;
arranging the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix, and then obtaining a historical weather feature matrix through a first convolution neural network model serving as a feature extractor;
arranging the power data of all preset time points in each day of the historical power data into power input vectors, and then obtaining single-day multi-scale power feature vectors corresponding to each day through the multi-scale neighborhood feature extraction module;
arranging the single-day multi-scale power characteristic vectors corresponding to each day into a two-dimensional characteristic matrix, and then obtaining a historical power characteristic matrix through a second convolution neural network serving as a characteristic extractor;
fusing the historical meteorological feature matrix and the historical power feature matrix to obtain a fused feature matrix;
acquiring meteorological data of a plurality of preset time points of the day to be predicted and output power of a wind power plant to be predicted at the preset time points;
converting meteorological data of a plurality of preset time points on the day to be predicted into meteorological feature vectors to be predicted, converting output power of the wind power plant to be predicted at the preset time points into power feature vectors to be predicted, and fusing the meteorological feature vectors to be predicted and the power feature vectors to be predicted to obtain predicted feature vectors;
multiplying the predicted eigenvector by the fused eigenvector matrix to obtain a decoded eigenvector;
correcting the characteristic values of all positions in the decoding characteristic vector based on the mean value and the variance of the characteristic value sets of all the positions of the decoding characteristic vector to obtain a corrected decoding characteristic vector;
and performing decoding regression on the corrected decoded feature vector through a decoder to obtain a decoded value, wherein the decoded value is used for representing the power predicted value of the wind power plant to be predicted at the next preset time point.
In the method for predicting short-time wind power for a wind farm, after arranging various data in meteorological data of each predetermined time point in historical meteorological data as input vectors, obtaining single-point multi-dimensional meteorological feature vectors through a full connection layer, the method includes: arranging various data in the meteorological data of each preset time point in the historical meteorological data into an input vector according to a time dimension; and performing full-connection coding on the input vector by using the full-connection layer according to the following formula to extract high-dimensional implicit features of feature values of all positions in the input vector so as to obtain the single-point multi-dimensional meteorological feature vector, wherein the formula is as follows:wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the offset vector,representing a matrix multiplication.
In the short-time wind power prediction method for the wind farm, the step of arranging the single-point multi-dimensional meteorological feature vectors into one-dimensional feature vectors with a single day as a time dimension and then obtaining the multi-scale single-weather image feature vectors corresponding to each day through the multi-scale neighborhood feature extraction module comprises the following steps: inputting the one-dimensional feature vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale single weather image feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; inputting the one-dimensional feature vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale single weather image feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and cascading the first neighborhood scale single weather image feature vector and the second neighborhood scale single weather image feature vector to obtain the multi-scale single weather image feature vector corresponding to each day.
In the short-time wind power prediction method for a wind farm, inputting the one-dimensional feature vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale single weather image feature vector, including: performing one-dimensional convolution coding on the one-dimensional feature vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first neighborhood scale single weather image feature vector;
wherein the formula is:
wherein a is the width of the first convolution kernel in the X direction, F is a first convolution kernel parameter vector, G is a local vector matrix operated with a convolution kernel function, w is the size of the first convolution kernel, and X represents a tension input vector; inputting the one-dimensional feature vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale single weather image feature vector, comprising: performing one-dimensional convolution coding on the one-dimensional feature vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second neighborhood scale single weather image feature vector;
wherein the formula is:
wherein b is the width of the second convolution kernel in the X direction, F is a parameter vector of the second convolution kernel, G is a local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents a tension input vector.
In the method for predicting short-time wind power for a wind farm, the method for obtaining a historical meteorological feature matrix by arranging the multi-scale single-weather image feature vectors corresponding to each day into a two-dimensional feature matrix and then using a first convolution neural network model as a feature extractor includes: respectively performing input data in forward transmission of layers by using each layer of the first convolutional neural network model as the feature extractor: carrying out convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on local channel dimensions on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the first convolution neural network model as the feature extractor is the historical meteorological feature matrix, and the input of the first layer of the first convolution neural network model as the feature extractor is the two-dimensional feature matrix.
In the short-time wind power prediction method for the wind power plant, fusing the historical meteorological characteristic matrix and the historical power characteristic matrix to obtain a fused characteristic matrix, which includes: fusing the historical meteorological feature matrix and the historical power feature matrix according to the following formula to obtain a fused feature matrix;
wherein the formula is:
M 1 =M*M 2
wherein M is 1 Representing said historical meteorological features matrix, M 2 Representing the historical power signature matrix, M representsThe fused feature matrix.
In the short-time wind power prediction method for a wind farm, the correcting the eigenvalue of each position in the decoded eigenvector based on the mean and variance of the eigenvalue sets of all positions of the decoded eigenvector to obtain a corrected decoded eigenvector includes: based on the mean and variance of the feature value sets of all the positions of the decoding feature vector, correcting the feature values of all the positions in the decoding feature vector by the following formula to obtain the corrected decoding feature vector;
wherein the formula is:
wherein v is i Representing the feature values of the respective positions in the decoded feature vector, V representing the decoded feature vector, and μ and σ being feature sets V, respectively i E mean and variance of V, and L is the length of the decoded feature vector, α is the weight hyperparameter.
In the short-time wind power prediction method for a wind farm, the decoding regression of the corrected decoded feature vector by a decoder to obtain a decoded value includes: decoding the corrected decoded feature vector using a plurality of fully-connected layers of the decoder to perform a decoding regression using the following formula to obtain the decoded value, wherein the formula is:wherein X is the corrected decoded feature vector, Y is the decoded value, W is a weight matrix, B is an offset vector,representing the matrix multiplication, h (-) is the activation function.
Compared with the prior art, the short-time wind power prediction system and the method for the wind power plant adopt an artificial intelligence prediction technology, a convolutional neural network model is used as a feature extractor, deep mining is carried out on implicit dynamic change feature information of historical meteorological data and historical power data in a time dimension and the correlation features between the historical meteorological information and the historical power information, and then the power value of the wind power plant to be predicted at the next preset time point is predicted based on a feature retrieval mode of the historical data, so that the development and utilization of wind power are improved.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally indicate like parts or steps.
FIG. 1 is a diagram of an application scenario of a short-term wind power prediction system for a wind farm according to an embodiment of the present application.
FIG. 2 is a block diagram of a short-time wind power prediction system for a wind farm according to an embodiment of the present application.
FIG. 3 is a flowchart of a short-term wind power prediction method for a wind farm according to an embodiment of the present application.
FIG. 4 is a schematic architecture diagram of a short-time wind power prediction method for a wind farm according to an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Overview of a scene
As described above, wind power is widely developed and utilized worldwide as the most valuable renewable energy source. However, due to the randomness of the wind speed, the wind power has great fluctuation and uncertainty, which affects the stable operation of the power system, thereby limiting the absorption of the power system to the wind power.
According to the recent statistical report of the State energy agency, the method comprises the following steps: the installed wind power capacity of the inner Mongolia accounts for about 30% of the total energy of the region, but the capacity share of the wind turbine generator on the power grid is less than 2% of the total installed wind power capacity. To improve the development and utilization of wind power, the accuracy of wind power prediction needs to be further improved.
Some existing wind power prediction schemes for wind power plants, for example, short-time power prediction is performed by using historical meteorological information and historical power information, but the accuracy of power prediction is not high due to the fact that implicit mode features in the historical meteorological information, implicit mode features in the historical power information and associations between the historical meteorological information and the historical power information cannot be sufficiently mined by the adopted data statistical model. Therefore, an optimized short-term wind power prediction scheme for wind farms is desired.
At present, deep learning and neural networks have been widely applied in the fields of computer vision, natural language processing, speech signal processing, and the like. In addition, deep learning and neural networks also exhibit a level close to or even exceeding that of humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like.
In recent years, deep learning and development of neural networks provide new solutions and schemes for short-time wind power prediction of wind power plants.
Based on this, in the technical scheme of the application, it is expected that a convolutional neural network model is used as a feature extractor to perform deep mining on implicit dynamic change feature information of historical meteorological data and historical power data in a time dimension and correlation features between the historical meteorological information and the historical power information, and then the power of the wind power plant to be predicted at the next preset time point is predicted based on feature retrieval of the historical data, so that development and utilization of wind power are improved.
Specifically, in the technical solution of the present application, first, historical meteorological data and historical power data are obtained, where the meteorological data at each predetermined time point in the historical meteorological data includes wind speed, wind direction, temperature, humidity, and pressure. Accordingly, in one specific example of the present application, data sampling may be performed at a sampling frequency of 15min, and the wind direction may be represented by a sine value or a cosine value of an angle.
Then, aiming at the meteorological data of each time point of the historical meteorological data, arranging all items of data in the meteorological data of each preset time point in the historical meteorological data into an input vector, and then extracting high-dimensional implicit features of all items of data in the meteorological data of each preset time point in the historical meteorological data through a full connection layer, so that a single-point multi-dimensional meteorological feature vector is obtained.
Further, the single-point multi-dimensional meteorological eigenvectors are arranged into a one-dimensional eigenvector by taking a single day as a time dimension so as to facilitate subsequent feature mining. It should be understood that convolutional neural networks were originally models applied in the image domain, but the idea of local feature extraction can be applied to time series data analysis as well. For example, a time-series convolution structure with a convolution kernel size of 3, the convolution kernel is moved along the time dimension in the form of a sliding window for time-series data input, and outputs a weighted sum of the data within each time-series segment. Each convolution unit stacks a plurality of convolution kernels to output a multi-dimensional feature. The large convolution kernel can extract features from a large-scale time sequence neighborhood, wherein the influence of each item of numerical value in the neighborhood is smaller, so that the fluctuation of input data is weakened, and the influence of noise points on output features is reduced. However, the difference of numerical value changes is weakened by the large-scale convolution kernel, and the problem of excessive smoothness is easily caused, so that the output characteristics lose the discrimination capability. In contrast, small scale convolution kernels are better able to retain information in the input data, but are also more susceptible to interference from noise therein. Therefore, the convolution units with different sizes are combined to extract the features of different time sequence scales in consideration of the characteristics of convolution with different scales. And then completing feature fusion by adopting a feature splicing mode, thereby obtaining the multi-scale neighborhood features.
That is, specifically, in the technical solution of the present application, further, the one-dimensional feature vectors are respectively subjected to one-dimensional convolutional coding by using convolutional layers of one-dimensional convolutional kernels with different scales of the multi-scale neighborhood feature extraction module, and then the obtained feature vectors corresponding to the two one-dimensional convolutional kernels with different scales are cascaded to obtain the multi-scale single weather image feature vector corresponding to each day. Particularly, by the method, the output features comprise the smoothed features and the original input features, so that the loss of information is avoided, and the accuracy of subsequent classification is improved. In other examples of the present application, the multi-scale neighborhood feature extraction module may further include a greater number of one-dimensional convolution layers, which use one-dimensional convolution kernels with different lengths to perform neighborhood-related feature extraction with different scales, which is not limited by the present application.
After obtaining the multi-scale single weather image characteristic vector corresponding to each day, arranging the multi-scale single weather image characteristic vectors corresponding to each day into a two-dimensional characteristic matrix to integrate the multi-scale neighborhood characteristics of the weather image of each day, and performing deep excavation of implicit associated characteristics in a first convolution neural network model serving as a characteristic extractor to obtain a historical weather characteristic matrix with global weather associated characteristic information of each day.
Then, similarly, for the historical power data of all the predetermined time points in each day, after the power data of all the predetermined time points in each day of the historical power data are arranged as power input vectors, one-dimensional convolution coding is performed in the convolution layer with the one-dimensional convolution kernels of different scales of the multi-scale neighborhood feature extraction module, and then the obtained feature vectors corresponding to the two one-dimensional convolution kernels of different scales are cascaded to obtain the single-day multi-scale power feature vector corresponding to each day. Further, after the single-day multi-scale power feature vectors corresponding to each day are arranged into a two-dimensional feature matrix to integrate the power multi-scale neighborhood features of each day, deep mining of implicit associated features is performed in a second convolutional neural network serving as a feature extractor, so that a historical power feature matrix with global power associated feature information of each day is obtained.
It should be understood that, considering that the feature scale of the dynamically changing feature of the output power is different from that of the weather in the high-dimensional space, and the dynamically changing feature of the output power can be regarded as a responsive feature of the dynamically changing feature of the weather in the high-dimensional feature space, in order to better fuse the historical weather feature matrix and the historical power feature matrix, a transition matrix of the historical weather feature matrix relative to the historical power feature matrix is further calculated to obtain a fused feature matrix.
When the power value of the wind power plant on a certain day to be predicted at the next preset time point is predicted, firstly, meteorological data of a plurality of preset time points on the day to be predicted are obtained through various meteorological sensors, and the output power of the wind power plant to be predicted at the preset time points is collected through a power detector. Accordingly, in one particular example, in the process of collecting the meteorological data, wind speed, wind direction, temperature, humidity and pressure data may be collected by a wind speed sensor, a wind vane, a temperature sensor, a humidity sensor and a pressure sensor, respectively.
In this way, after the meteorological data of a plurality of preset time points on the day to be predicted are converted into the meteorological characteristic vector to be predicted and the output power of the wind power plant to be predicted at the preset time points is converted into the power characteristic vector to be predicted, the meteorological characteristic vector to be predicted and the power characteristic vector to be predicted are fused in the same way to obtain the predicted characteristic vector. Further, multiplying the predicted feature vector by the fused feature matrix to map the predicted feature vector into a high-dimensional feature space of the fused feature matrix, thereby obtaining a decoded feature vector.
It should be understood that, since the fused feature matrix is used as the fusion of the historical meteorological feature matrix and the historical power feature matrix, and the predicted feature vector is used as the fusion of the meteorological feature vector to be predicted and the power feature vector to be predicted, while providing a multi-dimensional information fusion representation, divergence of feature distribution is also caused by the dimension difference of information, which makes the problem of the decoding feature vector obtained by multiplying the predicted feature vector and the fused feature matrix in this respect worse, which may cause poor decoding convergence of the decoding feature vector, and affect training speed and decoding accuracy.
Therefore, the information statistics normalization of the adaptive example is performed on the decoded feature vector, for example denoted as V, that is:
μ and σ are feature sets v, respectively i E.g., the mean and variance of V, and L is the length of the decoded feature vector V, and α is a weight hyperparameter.
Here, the eigenvalue set of the decoding eigenvector V is used as intrinsic priori (intra) information of statistical characteristics of the adaptive example to perform dynamic generation type information normalization on a single eigenvalue, and meanwhile, the normalized mode length information of the eigenvalue set is used as a bias to support invariance description in the set distribution domain, thereby realizing the convergence optimization of the eigenvector shielding the disturbance distribution of the special example as much as possible, improving the decoding convergence of the decoding eigenvector, and further improving the accuracy of decoding prediction.
Based on this, the present application proposes a short-time wind power prediction system for a wind farm, comprising: the historical data acquisition module is used for acquiring historical meteorological data and historical power data, wherein the meteorological data of each preset time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure; the single-point meteorological data encoding module is used for arranging various data in the meteorological data of each preset time point in the historical meteorological data into input vectors and then obtaining single-point multi-dimensional meteorological feature vectors through a full connection layer; the single-point multi-dimensional meteorological feature vector is arranged into a one-dimensional feature vector by taking a single day as a time dimension, and then the multi-scale single-weather image feature vector corresponding to each day is obtained through the multi-scale neighborhood feature extraction module; the global meteorological data coding module is used for arranging the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix and then obtaining a historical meteorological feature matrix through a first convolution neural network model serving as a feature extractor; the single-day power data coding module is used for arranging the power data of all the preset time points in each day of the historical power data into a power input vector and then obtaining a single-day multi-scale power feature vector corresponding to each day through the multi-scale neighborhood feature extraction module; the global power data coding module is used for arranging the single-day multi-scale power characteristic vectors corresponding to each day into a two-dimensional characteristic matrix and then obtaining a historical power characteristic matrix through a second convolutional neural network serving as a characteristic extractor; the historical data feature fusion module is used for fusing the historical meteorological feature matrix and the historical power feature matrix to obtain a fusion feature matrix; the data acquisition module on the day to be predicted is used for acquiring meteorological data of a plurality of preset time points on the day to be predicted and output power of the wind power plant to be predicted at the preset time points; the to-be-predicted data encoding module is used for converting meteorological data of a plurality of preset time points on the day to be predicted into to-be-predicted meteorological feature vectors, converting output power of the wind power plant to be predicted at the preset time points into to-be-predicted power feature vectors, and fusing the to-be-predicted meteorological feature vectors and the to-be-predicted power feature vectors to obtain predicted feature vectors; the feature query module is used for multiplying the predicted feature vector and the fusion feature matrix to obtain a decoding feature vector; the correction module is used for correcting the characteristic values of all positions in the decoding characteristic vector based on the mean value and the variance of the characteristic value sets of all the positions of the decoding characteristic vector to obtain a corrected decoding characteristic vector; and the predicted value generation module is used for performing decoding regression on the corrected decoded characteristic vector through a decoder to obtain a decoded value, and the decoded value is used for representing the power predicted value of the wind power plant to be predicted at the next preset time point.
FIG. 1 illustrates an application scenario diagram of a short-time wind power prediction system for a wind farm according to an embodiment of the application. As shown in fig. 1, in this application scenario, first, historical meteorological data and historical power data are acquired through a cloud storage terminal (e.g., D as illustrated in fig. 1), wherein the meteorological data at each predetermined time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure, and meteorological data at a plurality of predetermined time points on the day to be predicted are acquired through each meteorological sensor (e.g., a wind speed sensor T1, a wind vane T2, a temperature sensor T3, a humidity sensor T4 and a pressure sensor T5 as illustrated in fig. 1) and output power of a wind farm (e.g., F as illustrated in fig. 1) to be predicted at the plurality of predetermined time points are acquired through a power detector (e.g., P as illustrated in fig. 1). Then, the obtained historical meteorological data and historical power data, and meteorological data of a plurality of preset time points of the day to be predicted and output power of the wind farm to be predicted at the preset time points are input into a server (for example, a server S as illustrated in FIG. 1) deployed with a short-time wind power prediction algorithm for the wind farm, wherein the server can process the historical meteorological data and the historical power data, the meteorological data of the plurality of preset time points of the day to be predicted and the output power of the wind farm to be predicted at the preset time points in the short-time wind power prediction algorithm for the wind farm to generate a decoding value representing the power prediction value of the wind farm to be predicted at the next preset time point.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary System
FIG. 2 illustrates a block diagram of a short-time wind power prediction system for a wind farm according to an embodiment of the present application. As shown in fig. 2, a short-term wind power prediction system 200 for a wind farm according to an embodiment of the present application includes: the historical data acquisition module 210 is configured to acquire historical meteorological data and historical power data, where the meteorological data at each predetermined time point in the historical meteorological data includes wind speed, wind direction, temperature, humidity, and pressure; the single-point meteorological data encoding module 220 is configured to arrange various data in the meteorological data of each predetermined time point in the historical meteorological data into input vectors and then obtain single-point multidimensional meteorological feature vectors through a full connection layer; the single-weather image data encoding module 230 is configured to arrange the single-point multi-dimensional weather feature vectors into one-dimensional feature vectors with a single day as a time dimension, and then obtain multi-scale single-weather image feature vectors corresponding to each day through the multi-scale neighborhood feature extraction module; the global meteorological data encoding module 240 is configured to arrange the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix, and then obtain a historical meteorological feature matrix through a first convolution neural network model serving as a feature extractor; the single-day power data encoding module 250 is configured to arrange the power data of all the predetermined time points in each day of the historical power data into a power input vector, and then obtain a single-day multi-scale power feature vector corresponding to each day through the multi-scale neighborhood feature extraction module; the global power data coding module 260 is used for arranging the single-day multi-scale power feature vectors corresponding to each day into a two-dimensional feature matrix and then obtaining a historical power feature matrix through a second convolutional neural network serving as a feature extractor; a historical data feature fusion module 270, configured to fuse the historical meteorological feature matrix and the historical power feature matrix to obtain a fusion feature matrix; the data acquisition module 280 on the day to be predicted is used for acquiring meteorological data of a plurality of preset time points on the day to be predicted and output power of the wind power plant to be predicted at the preset time points; the to-be-predicted data encoding module 290 is configured to convert the meteorological data at the multiple predetermined time points of the day to be predicted into a to-be-predicted meteorological eigenvector, convert the output power of the wind farm to be predicted at the multiple predetermined time points into a to-be-predicted power eigenvector, and then fuse the to-be-predicted meteorological eigenvector and the to-be-predicted power eigenvector to obtain a predicted eigenvector; a feature query module 300, configured to multiply the predicted feature vector and the fused feature matrix to obtain a decoded feature vector; a correcting module 310, configured to correct feature values of each position in the decoded feature vector based on a mean and a variance of feature value sets of all positions of the decoded feature vector to obtain a corrected decoded feature vector; and the predicted value generating module 320 is configured to perform decoding regression on the corrected decoded feature vector through a decoder to obtain a decoded value, where the decoded value is used to represent a power predicted value of the wind farm to be predicted at the next predetermined time point.
Specifically, in the embodiment of the present application, the historical data collecting module 210 and the single-point meteorological data encoding module 220 are configured to obtain historical meteorological data and historical power data, where the meteorological data at each predetermined time point in the historical meteorological data includes wind speed, wind direction, temperature, humidity, and pressure, and arrange each item of data in the meteorological data at each predetermined time point in the historical meteorological data as an input vector and then pass through a full connection layer to obtain a single-point multidimensional meteorological feature vector. As described above, in the technical solution of the present application, it is desirable to use a convolutional neural network model as a feature extractor to perform deep mining on implicit dynamic change feature information of historical meteorological data and historical power data in a time dimension and associated features between the historical meteorological information and the historical power information, and then predict the power of the wind farm to be predicted at the next predetermined time point based on feature retrieval of the historical data, thereby improving development and utilization of wind power.
That is, specifically, in the technical solution of the present application, first, historical meteorological data and historical power data are obtained, wherein the meteorological data at each predetermined time point in the historical meteorological data include wind speed, wind direction, temperature, humidity and pressure. Accordingly, in one specific example of the present application, data sampling may be performed at a sampling frequency of 15min, and the wind direction may be represented by a sine value or a cosine value of an angle. Then, aiming at the meteorological data of each time point of the historical meteorological data, arranging all items of data in the meteorological data of each preset time point in the historical meteorological data into an input vector, and then extracting high-dimensional implicit features of all items of data in the meteorological data of each preset time point in the historical meteorological data through a full connection layer, so that a single-point multi-dimensional meteorological feature vector is obtained.
More specifically, in this embodiment of the present application, the single-point weather data encoding module is further configured to: arranging various data in the meteorological data of each preset time point in the historical meteorological data into an input vector according to a time dimension; and performing full-connection coding on the input vector by using the full-connection layer according to the following formula to extract high-dimensional implicit features of feature values of all positions in the input vector so as to obtain the single-point multi-dimensional meteorological feature vector, wherein the formula is as follows: wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the offset vector,representing a matrix multiplication.
Specifically, in this embodiment of the present application, the single-point multi-dimensional weather image data encoding module 230 is configured to arrange the single-point multi-dimensional weather feature vectors into one-dimensional feature vectors with a single day as a time dimension, and then obtain the multi-scale single weather image feature vectors corresponding to each day through the multi-scale neighborhood feature extraction module. That is, in the technical solution of the present application, further, the single-point multidimensional meteorological eigenvector is arranged as a one-dimensional eigenvector with a single day as a time dimension, so as to facilitate subsequent feature mining. It should be understood that convolutional neural networks were originally models applied in the image domain, but the idea of local feature extraction can be applied to time series data analysis as well. For example, a time-series convolution structure with a convolution kernel size of 3, the convolution kernel is moved along the time dimension in the form of a sliding window for time-series data input, and outputs a weighted sum of the data within each time-series segment. Each convolution unit stacks a plurality of convolution kernels to output a multi-dimensional feature. The large convolution kernel can extract features from a large-scale time sequence neighborhood, wherein the influence of each item of numerical value in the neighborhood is smaller, so that the fluctuation of input data is weakened, and the influence of noise points on output features is reduced. However, the difference of numerical value changes is weakened by the large-scale convolution kernel, and the problem of excessive smoothness is easily caused, so that the output characteristics lose the discrimination capability. In contrast, small scale convolution kernels are better able to retain information in the input data, but are also more susceptible to interference from noise therein. Therefore, the convolution units with different sizes are combined to extract the features of different time sequence scales in consideration of the characteristics of convolution with different scales. And then completing feature fusion by adopting a feature splicing mode, thereby obtaining the multi-scale neighborhood features.
That is to say, specifically, in the technical solution of the present application, further, the convolution layers of the one-dimensional convolution kernels with different scales of the multi-scale neighborhood feature extraction module are used to respectively perform one-dimensional convolution encoding on the one-dimensional feature vectors, and then the obtained feature vectors corresponding to the two one-dimensional convolution kernels with different scales are cascaded to obtain the multi-scale single weather image feature vector corresponding to each day. Particularly, by the method, the output features comprise the smoothed features and the original input features, so that the loss of information is avoided, and the accuracy of subsequent classification is improved. In other examples of the present application, the multi-scale neighborhood feature extraction module may further include a greater number of one-dimensional convolution layers, which perform neighborhood-related feature extraction of different scales using one-dimensional convolution kernels of different lengths, which is not limited in this application.
More specifically, in this embodiment of the present application, the single weather image data encoding module includes: the first convolution unit is used for inputting the one-dimensional feature vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale single weather image feature vector, wherein the first convolution layer is provided with a first one-dimensional convolution kernel with a first length. Accordingly, in a specific example, the one-dimensional feature vector is subjected to one-dimensional convolution coding by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first neighborhood scale single weather image feature vector;
wherein the formula is:
wherein, a is the width of the first convolution kernel in the X direction, F is a first convolution kernel parameter vector, G is a local vector matrix operated with the convolution kernel function, w is the size of the first convolution kernel, and X represents a tension input vector. And the second convolution unit is used for inputting the one-dimensional feature vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale single weather image feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length. Accordingly, in one specific example, the one-dimensional feature vector is subjected to one-dimensional convolution encoding by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second neighborhood scale single weather image feature vector;
wherein the formula is:
wherein b is the width of the second convolution kernel in the X direction, F is a parameter vector of the second convolution kernel, G is a local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents a tension input vector. The cascade unit is used for cascading the first neighborhood scale single weather image feature vector and the second neighborhood scale single weather image feature vector to obtain the multi-scale single weather image feature vector corresponding to each day.
Specifically, in this embodiment of the present application, the global weather data encoding module 240 is configured to arrange the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix, and then obtain a historical weather feature matrix through a first convolution neural network model serving as a feature extractor. That is, in the technical solution of the present application, after obtaining the multi-scale single weather image feature vector corresponding to each day, further arranging the multi-scale single weather image feature vector corresponding to each day into a two-dimensional feature matrix to integrate the multi-scale neighborhood features of the weather image on each day, and then performing deep mining of implicit associated features in the first convolution neural network model serving as the feature extractor to obtain the historical weather feature matrix with global weather associated feature information on each day.
More specifically, in this embodiment of the application, the global weather data encoding module is further configured to: using each layer of the first convolutional neural network model as a feature extractor to respectively perform the following steps on input data in the forward direction transfer of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on local channel dimensions on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the first convolution neural network model as the feature extractor is the historical meteorological feature matrix, and the input of the first layer of the first convolution neural network model as the feature extractor is the two-dimensional feature matrix.
Specifically, in this embodiment of the present application, the single-day power data encoding module 250 and the global power data encoding module 260 are configured to arrange the power data of all predetermined time points in each day of the historical power data into a power input vector, then pass through the multi-scale neighborhood feature extraction module to obtain a single-day multi-scale power feature vector corresponding to each day, and arrange the single-day multi-scale power feature vector corresponding to each day into a two-dimensional feature matrix, and then pass through a second convolutional neural network serving as a feature extractor to obtain a historical power feature matrix. That is, in the technical solution of the present application, similarly, for the historical power data of all the predetermined time points in each day, after the power data of all the predetermined time points in each day of the historical power data are arranged as power input vectors, one-dimensional convolution coding is performed in the convolution layer of the one-dimensional convolution kernels with different scales of the multi-scale neighborhood feature extraction module, and then the obtained feature vectors corresponding to the two one-dimensional convolution kernels with different scales are concatenated to obtain the single-day multi-scale power feature vector corresponding to each day.
Further, after the single-day multi-scale power feature vectors corresponding to each day are arranged into a two-dimensional feature matrix to integrate the power multi-scale neighborhood features of each day, deep mining of implicit associated features is performed in a second convolutional neural network serving as a feature extractor, so that a historical power feature matrix with global power associated feature information of each day is obtained.
Specifically, in this embodiment of the present application, the historical data feature fusion module 270 is configured to fuse the historical meteorological feature matrix and the historical power feature matrix to obtain a fusion feature matrix. It should be understood that, considering that the feature scale of the dynamically changing feature of the output power is different from that of the weather in the high-dimensional space, and the dynamically changing feature of the output power can be regarded as a responsive feature of the dynamically changing feature of the weather in the high-dimensional feature space, in order to better fuse the historical weather feature matrix and the historical power feature matrix, a transition matrix of the historical weather feature matrix relative to the historical power feature matrix is further calculated to obtain a fused feature matrix.
More specifically, in this embodiment of the present application, the historical data feature fusion module is further configured to: fusing the historical meteorological feature matrix and the historical power feature matrix according to the following formula to obtain a fused feature matrix;
wherein the formula is:
M 1 =M*M 2
wherein M is 1 Representing said historical meteorological features matrix, M 2 Representing the historical power feature matrix, and M representing the fusion feature matrix.
Specifically, in the embodiment of the present application, the data acquisition module 280 on the day to be predicted is configured to obtain meteorological data of a plurality of predetermined time points on the day to be predicted and output power of a wind farm to be predicted at the plurality of predetermined time points. That is, in the technical solution of the present application, when a power value of a wind farm on a certain day to be predicted at a next predetermined time point is predicted, firstly, meteorological data of a plurality of predetermined time points on the day to be predicted are acquired by each meteorological sensor, and output power of the wind farm to be predicted at the plurality of predetermined time points is acquired by a power detector. Accordingly, in one particular example, in the process of collecting the meteorological data, wind speed, wind direction, temperature, humidity and pressure data may be collected by a wind speed sensor, a wind vane, a temperature sensor, a humidity sensor and a pressure sensor, respectively.
Specifically, in this embodiment of the present application, the to-be-predicted data encoding module 290 and the feature query module 300 are configured to convert the meteorological data at multiple predetermined time points on the day to be predicted into a to-be-predicted meteorological feature vector, convert the output power of the wind farm to be predicted at the multiple predetermined time points into a to-be-predicted power feature vector, fuse the to-be-predicted meteorological feature vector and the to-be-predicted power feature vector to obtain a predicted feature vector, and multiply the predicted feature vector and the fused feature matrix to obtain a decoded feature vector. That is to say, in the technical scheme of the application, after meteorological data at a plurality of predetermined time points on the day to be predicted are converted into meteorological eigenvectors to be predicted and output power of the wind farm to be predicted at the plurality of predetermined time points is converted into power eigenvectors to be predicted, the meteorological eigenvectors to be predicted and the power eigenvectors to be predicted are fused in the same manner to obtain predicted eigenvectors. Further, multiplying the predicted feature vector and the fused feature matrix to map the predicted feature vector into a high-dimensional feature space of the fused feature matrix, thereby obtaining a decoded feature vector.
Specifically, in this embodiment of the present application, the correcting module 310 is configured to correct the eigenvalue of each position in the decoded eigenvector based on a mean and a variance of the eigenvalue set of all positions of the decoded eigenvector to obtain a corrected decoded eigenvector. It should be understood that, since the fused feature matrix is used as the fusion of the historical meteorological feature matrix and the historical power feature matrix, and the predicted feature vector is used as the fusion of the meteorological feature vector to be predicted and the power feature vector to be predicted, while providing a multi-dimensional information fusion representation, divergence of feature distribution is also caused by the dimension difference of information, which makes the problem of the decoding feature vector obtained by multiplying the predicted feature vector and the fused feature matrix in this respect worse, which may cause poor decoding convergence of the decoding feature vector, and affect training speed and decoding accuracy. Therefore, in the technical solution of the present application, the decoding feature vector, for example, denoted as V, is further subjected to statistical information normalization of the adaptive example.
That is, here, the eigenvalue set of the decoding eigenvector V is used as intrinsic prior (intra _ priors) information of statistical characteristics of the adaptive example to perform dynamic generation type information normalization on a single eigenvalue, and meanwhile, the normalization mode length information of the eigenvalue set is used as a bias to support invariance description in the set distribution domain, thereby realizing convergence optimization of the eigenvector shielding disturbance distribution of the special example as much as possible, thereby improving the decoding convergence of the decoding eigenvector and further improving the accuracy of decoding prediction.
More specifically, in this embodiment of the application, the correction module is further configured to: based on the mean and variance of the feature value sets of all the positions of the decoding feature vector, correcting the feature values of all the positions in the decoding feature vector by the following formula to obtain the corrected decoding feature vector;
wherein the formula is:
wherein v is i Representing the eigenvalues of the respective positions in the decoded eigenvector, V representing the decoded eigenvector, mu and sigma being the sets of features V, respectively i E.g., the mean and variance of V, and L is the length of the decoded feature vector, and α is a weight hyperparameter.
Specifically, in this embodiment of the present application, the predicted value generating module 320 is configured to perform decoding regression on the corrected decoded feature vector through a decoder to obtain a decoded value, where the decoded value is used to indicate a power predicted value of the wind farm to be predicted at a next predetermined time point. Accordingly, in one specific example, the corrected decoded feature vector is subjected to decoding regression using a plurality of fully-connected layers of the decoder to obtain the decoded value according to the following formula:wherein X is the corrected decoded feature vector, Y is the decoded value, W is a weight matrix, B is an offset vector,representing the matrix multiplication, h (-) is the activation function.
In summary, the short-time wind power prediction system 200 for a wind farm based on the embodiment of the present application is illustrated, and an artificial intelligence prediction technology is adopted, and a convolutional neural network model is used as a feature extractor to perform deep mining on implicit dynamic change feature information of historical meteorological data and historical power data in a time dimension and correlation features between the historical meteorological information and the historical power information, so as to predict a power value of the wind farm to be predicted at a next predetermined time point based on a feature retrieval manner of the historical data, so as to improve development and utilization of wind power.
As described above, the short-term wind power prediction system 200 for a wind farm according to the embodiment of the present application may be implemented in various terminal devices, such as a server of a short-term wind power prediction algorithm for a wind farm, and the like. In one example, the short-term wind power prediction system 200 for a wind farm according to embodiments of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the short-term wind power prediction system 200 for a wind farm may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the short-term wind power prediction system 200 for a wind farm may also be one of many hardware modules of the terminal device.
Alternatively, in another example, the short-term wind power prediction system for wind farm 200 and the terminal device may also be separate devices, and the short-term wind power prediction system for wind farm 200 may be connected to the terminal device through a wired and/or wireless network and transmit the interaction information according to an agreed data format.
Exemplary method
FIG. 3 illustrates a flow chart of a short-time wind power prediction method for a wind farm. As shown in FIG. 3, the short-time wind power prediction method for the wind farm according to the embodiment of the application comprises the following steps: s110, acquiring historical meteorological data and historical power data, wherein the meteorological data of each preset time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure; s120, arranging all data in the meteorological data of all the preset time points in the historical meteorological data into input vectors, and then obtaining single-point multi-dimensional meteorological feature vectors through a full connection layer; s130, arranging the single-point multi-dimensional meteorological feature vectors into one-dimensional feature vectors by taking a single day as a time dimension, and then obtaining multi-scale single-meteorological feature vectors corresponding to each day through a multi-scale neighborhood feature extraction module; s140, arranging the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix, and then obtaining a historical weather feature matrix through a first convolution neural network model serving as a feature extractor; s150, arranging the power data of all the preset time points in each day of the historical power data into power input vectors, and then obtaining single-day multi-scale power feature vectors corresponding to each day through the multi-scale neighborhood feature extraction module; s160, arranging the single-day multi-scale power characteristic vectors corresponding to each day into a two-dimensional characteristic matrix, and then obtaining a historical power characteristic matrix through a second convolution neural network serving as a characteristic extractor; s170, fusing the historical meteorological feature matrix and the historical power feature matrix to obtain a fused feature matrix; s180, acquiring meteorological data of a plurality of preset time points of the day to be predicted and output power of the wind power plant to be predicted at the preset time points; s190, converting meteorological data of a plurality of preset time points on the day to be predicted into meteorological feature vectors to be predicted, converting output power of the wind power plant to be predicted at the preset time points into power feature vectors to be predicted, and fusing the meteorological feature vectors to be predicted and the power feature vectors to be predicted to obtain predicted feature vectors; s200, multiplying the predicted feature vector and the fusion feature matrix to obtain a decoding feature vector; s210, correcting the characteristic values of all positions in the decoding characteristic vector based on the mean value and the variance of the characteristic value sets of all the positions of the decoding characteristic vector to obtain a corrected decoding characteristic vector; and S220, performing decoding regression on the corrected decoding characteristic vector through a decoder to obtain a decoding value, wherein the decoding value is used for representing the power prediction value of the wind power plant to be predicted at the next preset time point.
FIG. 4 illustrates an architectural diagram of a short-time wind power prediction method for a wind farm according to an embodiment of the application. As shown in fig. 4, in the network architecture of the short-time wind power prediction method for a wind farm, first, after arranging items of data in meteorological data of various predetermined time points in the obtained historical meteorological data (e.g., P1 as illustrated in fig. 4) as an input vector (e.g., V1 as illustrated in fig. 4), a single-point multidimensional meteorological feature vector (e.g., VF1 as illustrated in fig. 4) is obtained through a fully-connected layer (e.g., FC as illustrated in fig. 4); then, arranging the single-point multi-dimensional meteorological feature vectors into one-dimensional feature vectors (for example, VF2 as illustrated in fig. 4) by taking a single day as a time dimension, and then passing through a multi-scale neighborhood feature extraction module (for example, MS as illustrated in fig. 4) to obtain multi-scale single-dimensional meteorological feature vectors (for example, VF3 as illustrated in fig. 4) corresponding to each day; then, arranging the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix (for example, MF1 as illustrated in fig. 4) and then obtaining a historical weather feature matrix (for example, MF2 as illustrated in fig. 4) through a first convolution neural network model (for example, CNN1 as illustrated in fig. 4) serving as a feature extractor; then, arranging the power data of all the preset time points in each day of the obtained historical power data (for example, P2 as illustrated in FIG. 4) into a power input vector (for example, V2 as illustrated in FIG. 4) and then passing through the multi-scale neighborhood feature extraction module to obtain a single-day multi-scale power feature vector (for example, VF4 as illustrated in FIG. 4) corresponding to each day; then, arranging the single-day multi-scale power feature vectors corresponding to each day into a two-dimensional feature matrix (for example, MF3 as illustrated in fig. 4) and then passing through a second convolutional neural network (for example, CNN2 as illustrated in fig. 4) as a feature extractor to obtain a historical power feature matrix (for example, MF4 as illustrated in fig. 4); then, fusing the historical meteorological feature matrix and the historical power feature matrix to obtain a fused feature matrix (e.g., MF5 as illustrated in FIG. 4); then, after converting the obtained meteorological data (for example, Q1 as illustrated in fig. 4) of a plurality of predetermined time points of the day to be predicted into a meteorological feature vector to be predicted (for example, VM as illustrated in fig. 4) and converting the obtained output power (for example, Q2 as illustrated in fig. 4) of the wind farm to be predicted at the plurality of predetermined time points into a power feature vector to be predicted (for example, VP as illustrated in fig. 4), fusing the meteorological feature vector to be predicted and the power feature vector to be predicted to obtain a predicted feature vector (for example, VF as illustrated in fig. 4); then, multiplying the prediction feature vector and the fusion feature matrix to obtain a decoding feature vector (for example, VC as illustrated in fig. 4); then, based on the mean and variance of the feature value sets of all positions of the decoded feature vector, correcting the feature value of each position in the decoded feature vector to obtain a corrected decoded feature vector (e.g., VR as illustrated in fig. 4); finally, the corrected decoded feature vector is decoded and regressed by a decoder (for example, as shown in fig. 4 as D) to obtain a decoded value, wherein the decoded value is used for representing the power predicted value of the wind farm to be predicted at the next predetermined time point.
More specifically, in step S110 and step S120, historical meteorological data and historical power data are obtained, wherein the meteorological data at each predetermined time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure, and the meteorological data at each predetermined time point in the historical meteorological data are arranged into an input vector and then pass through a full connection layer to obtain a single-point multi-dimensional meteorological feature vector. That is, in the technical scheme of the application, it is desirable to use a convolutional neural network model as a feature extractor to perform deep mining on implicit dynamic change feature information of historical meteorological data and historical power data in a time dimension and associated features between the historical meteorological information and the historical power information, and then predict the power of the wind farm to be predicted at the next predetermined time point based on feature retrieval of the historical data, thereby improving development and utilization of wind power.
That is, specifically, in the technical solution of the present application, first, historical meteorological data and historical power data are obtained, wherein the meteorological data at each predetermined time point in the historical meteorological data include wind speed, wind direction, temperature, humidity and pressure. Accordingly, in one specific example of the present application, data sampling may be performed at a sampling frequency of 15min, and the wind direction may be represented by a sine value or a cosine value of an angle. Then, aiming at the meteorological data of each time point of the historical meteorological data, arranging all items of data in the meteorological data of each preset time point in the historical meteorological data into an input vector, and then extracting high-dimensional implicit features of all items of data in the meteorological data of each preset time point in the historical meteorological data through a full connection layer, so that a single-point multi-dimensional meteorological feature vector is obtained.
More specifically, in step S130, the single-point multi-dimensional meteorological feature vectors are arranged into one-dimensional feature vectors with a single day as a time dimension, and then the one-dimensional feature vectors are passed through a multi-scale neighborhood feature extraction module to obtain multi-scale single-meteorological feature vectors corresponding to each day. That is, in the technical solution of the present application, further, the single-point multidimensional meteorological eigenvector is arranged as a one-dimensional eigenvector with a single day as a time dimension, so as to facilitate subsequent feature mining. It should be understood that convolutional neural networks were originally models applied in the image domain, but the idea of local feature extraction can be applied to time series data analysis as well. For example, a time-series convolution structure with a convolution kernel size of 3, the convolution kernel is moved along the time dimension in the form of a sliding window for time-series data input, and outputs a weighted sum of the data within each time-series segment. Each convolution unit stacks a plurality of convolution kernels to output a multi-dimensional feature. The large convolution kernel can extract features from a large-scale time sequence neighborhood, wherein the influence of each item of value in the neighborhood is smaller, so that the fluctuation of input data is weakened, and the influence of noise points on output features is relieved. However, the difference of numerical value changes is weakened by the large-scale convolution kernel, and the problem of excessive smoothness is easily caused, so that the output characteristics lose the discrimination capability. In contrast, small scale convolution kernels are better able to retain information in the input data, but are also more susceptible to interference from noise therein. Therefore, the convolution units with different sizes are combined to extract the features of different time sequence scales in consideration of the characteristics of convolution with different scales. And then completing feature fusion by adopting a feature splicing mode, thereby obtaining the multi-scale neighborhood features.
That is, specifically, in the technical solution of the present application, further, the one-dimensional feature vectors are respectively subjected to one-dimensional convolutional coding by using convolutional layers of one-dimensional convolutional kernels with different scales of the multi-scale neighborhood feature extraction module, and then the obtained feature vectors corresponding to the two one-dimensional convolutional kernels with different scales are cascaded to obtain the multi-scale single weather image feature vector corresponding to each day. Particularly, by the method, the output features comprise the smoothed features and the original input features, so that the loss of information is avoided, and the accuracy of subsequent classification is improved. In other examples of the present application, the multi-scale neighborhood feature extraction module may further include a greater number of one-dimensional convolution layers, which perform neighborhood-related feature extraction of different scales using one-dimensional convolution kernels of different lengths, which is not limited in this application.
More specifically, in step S140, the multi-scale single weather image feature vectors corresponding to each day are arranged into a two-dimensional feature matrix, and then a historical weather feature matrix is obtained through a first convolution neural network model serving as a feature extractor. That is, in the technical solution of the present application, after obtaining the multi-scale single weather image feature vector corresponding to each day, further arranging the multi-scale single weather image feature vector corresponding to each day into a two-dimensional feature matrix to integrate the multi-scale neighborhood features of the weather image of each day, and then performing deep mining of implicit association features in the first convolution neural network model serving as the feature extractor to obtain the historical weather feature matrix with global weather association feature information of each day.
More specifically, in step S150 and step S160, after the power data of all the predetermined time points in each day of the historical power data are arranged as a power input vector, the multi-scale neighborhood feature extraction module obtains a single-day multi-scale power feature vector corresponding to each day, and after the single-day multi-scale power feature vector corresponding to each day is arranged as a two-dimensional feature matrix, the second convolutional neural network serving as a feature extractor obtains a historical power feature matrix. That is, in the technical solution of the present application, similarly, for the historical power data of all the predetermined time points in each day, after the power data of all the predetermined time points in each day of the historical power data are arranged as power input vectors, one-dimensional convolution coding is performed in the convolution layer of the one-dimensional convolution kernels with different scales of the multi-scale neighborhood feature extraction module, and then the obtained feature vectors corresponding to the two one-dimensional convolution kernels with different scales are concatenated to obtain the single-day multi-scale power feature vector corresponding to each day.
Further, after the single-day multi-scale power feature vectors corresponding to each day are arranged into a two-dimensional feature matrix to integrate the power multi-scale neighborhood features of each day, deep mining of implicit associated features is performed in a second convolutional neural network serving as a feature extractor, so that a historical power feature matrix with global power associated feature information of each day is obtained.
More specifically, in step S170, the historical meteorological feature matrix and the historical power feature matrix are fused to obtain a fused feature matrix. It should be understood that, considering that the feature scale of the dynamically changing feature of the output power is different from that of the weather in the high-dimensional space, and the dynamically changing feature of the output power can be regarded as a responsive feature of the dynamically changing feature of the weather in the high-dimensional feature space, in order to better fuse the historical weather feature matrix and the historical power feature matrix, a transition matrix of the historical weather feature matrix relative to the historical power feature matrix is further calculated to obtain a fused feature matrix.
More specifically, in step S180, meteorological data of a plurality of predetermined time points on the day to be predicted and output power of the wind farm to be predicted at the plurality of predetermined time points are acquired. That is, in the technical solution of the present application, when a power value of a wind farm on a certain day to be predicted at a next predetermined time point is predicted, firstly, meteorological data of a plurality of predetermined time points on the day to be predicted are acquired by each meteorological sensor, and output power of the wind farm to be predicted at the plurality of predetermined time points is acquired by a power detector. Accordingly, in one particular example, in the process of collecting the meteorological data, wind speed, wind direction, temperature, humidity and pressure data may be collected by a wind speed sensor, a wind vane, a temperature sensor, a humidity sensor and a pressure sensor, respectively.
More specifically, in step S190 and step S200, after converting meteorological data at a plurality of predetermined time points on the day to be predicted into meteorological feature vectors to be predicted and converting output power of the wind farm to be predicted at the plurality of predetermined time points into power feature vectors to be predicted, fusing the meteorological feature vectors to be predicted and the power feature vectors to be predicted to obtain predicted feature vectors, and multiplying the predicted feature vectors and the fused feature matrix to obtain decoded feature vectors. That is, in the technical solution of the present application, after converting the meteorological data at a plurality of predetermined time points on the day to be predicted into the meteorological feature vector to be predicted and converting the output power of the wind farm to be predicted at the plurality of predetermined time points into the power feature vector to be predicted, the meteorological feature vector to be predicted and the power feature vector to be predicted are fused in the same manner to obtain the predicted feature vector. Further, multiplying the predicted feature vector and the fused feature matrix to map the predicted feature vector into a high-dimensional feature space of the fused feature matrix, thereby obtaining a decoded feature vector.
More specifically, in step S210, the feature values of the positions in the decoded feature vector are corrected based on the mean and variance of the feature value sets of all the positions of the decoded feature vector to obtain a corrected decoded feature vector. It should be understood that, since the fused feature matrix is used as the fusion of the historical meteorological feature matrix and the historical power feature matrix, and the predicted feature vector is used as the fusion of the meteorological feature vector to be predicted and the power feature vector to be predicted, while providing a multi-dimensional information fusion representation, divergence of feature distribution is also caused by the dimension difference of information, which makes the problem of the decoding feature vector obtained by multiplying the predicted feature vector and the fused feature matrix in this respect worse, which may cause poor decoding convergence of the decoding feature vector, and affect training speed and decoding accuracy. Therefore, in the technical solution of the present application, the decoding feature vector, for example, denoted as V, is further subjected to statistical information normalization of the adaptive example.
That is, here, the eigenvalue set of the decoding eigenvector V is used as intrinsic prior (intra _ priors) information of statistical characteristics of the adaptive example to perform dynamic generation type information normalization on a single eigenvalue, and meanwhile, the normalization mode length information of the eigenvalue set is used as a bias to support invariance description in the set distribution domain, thereby realizing convergence optimization of the eigenvector shielding disturbance distribution of the special example as much as possible, thereby improving the decoding convergence of the decoding eigenvector and further improving the accuracy of decoding prediction.
More specifically, in step S220, the corrected decoded feature vector is subjected to decoding regression through a decoder to obtain a decoded value, where the decoded value is used to represent a power predicted value of the wind farm to be predicted at the next predetermined time point.
In summary, the short-time wind power prediction method for the wind farm based on the embodiment of the present application is elucidated, and an artificial intelligence prediction technology is adopted, and a convolutional neural network model is used as a feature extractor to perform deep mining on implicit dynamic change feature information of historical meteorological data and historical power data in a time dimension and correlation features between the historical meteorological information and the historical power information, so as to predict a power value of the wind farm to be predicted at a next predetermined time point based on a feature retrieval mode of the historical data, so as to improve development and utilization of wind power.
The foregoing describes the general principles of the present application in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present application are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.
The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably herein. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".
It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations are to be considered as equivalents of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.
Claims (10)
1. A short-term wind power prediction system for a wind farm, comprising:
the historical data acquisition module is used for acquiring historical meteorological data and historical power data, wherein the meteorological data of each preset time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure;
the single-point meteorological data encoding module is used for arranging various data in meteorological data of each preset time point in the historical meteorological data into input vectors and then obtaining single-point multi-dimensional meteorological characteristic vectors through a full connection layer;
the single-point multi-dimensional meteorological feature vector is arranged into a one-dimensional feature vector by taking a single day as a time dimension, and then the multi-scale single-weather image feature vector corresponding to each day is obtained through the multi-scale neighborhood feature extraction module;
the global meteorological data coding module is used for arranging the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix and then obtaining a historical meteorological feature matrix through a first convolution neural network model serving as a feature extractor;
the single-day power data coding module is used for arranging the power data of all the preset time points in each day of the historical power data into power input vectors and then obtaining single-day multi-scale power characteristic vectors corresponding to each day through the multi-scale neighborhood characteristic extraction module;
the global power data coding module is used for arranging the single-day multi-scale power characteristic vectors corresponding to each day into a two-dimensional characteristic matrix and then obtaining a historical power characteristic matrix through a second convolutional neural network serving as a characteristic extractor;
the historical data feature fusion module is used for fusing the historical meteorological feature matrix and the historical power feature matrix to obtain a fusion feature matrix;
the data acquisition module on the day to be predicted is used for acquiring meteorological data of a plurality of preset time points on the day to be predicted and output power of the wind power plant to be predicted at the preset time points;
the to-be-predicted data encoding module is used for converting the meteorological data of a plurality of preset time points of the day to be predicted into to-be-predicted meteorological characteristic vectors, converting the output power of the wind power plant to be predicted at the preset time points into to-be-predicted power characteristic vectors, and then fusing the to-be-predicted meteorological characteristic vectors and the to-be-predicted power characteristic vectors to obtain predicted characteristic vectors; and
the feature query module is used for multiplying the predicted feature vector and the fusion feature matrix to obtain a decoding feature vector;
the correction module is used for correcting the characteristic values of all positions in the decoding characteristic vector based on the mean value and the variance of the characteristic value sets of all the positions of the decoding characteristic vector to obtain a corrected decoding characteristic vector;
and the predicted value generation module is used for performing decoding regression on the corrected decoded characteristic vector through a decoder to obtain a decoded value, and the decoded value is used for representing the power predicted value of the wind power plant to be predicted at the next preset time point.
2. The short-time wind power prediction system for a wind farm of claim 1, wherein the single point meteorological data encoding module is further configured to: arranging various data in the meteorological data of each preset time point in the historical meteorological data into an input vector according to a time dimension; and performing full-connection coding on the input vector by using the full-connection layer according to the following formula to extract high-dimensional implicit features of feature values of all positions in the input vector so as to obtain the single-point multi-dimensional meteorological feature vector, wherein the formula is as follows:wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the offset vector,representing a matrix multiplication.
3. The short-time wind power prediction system for wind farms according to claim 2, characterized in that the single weather image data encoding module comprises:
the first convolution unit is used for inputting the one-dimensional feature vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale single weather image feature vector, wherein the first convolution layer is provided with a first one-dimensional convolution kernel with a first length;
a second convolution unit, configured to input the one-dimensional feature vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale single weather image feature vector, where the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and
the cascade unit is used for cascading the first neighborhood scale single weather image feature vector and the second neighborhood scale single weather image feature vector to obtain the multi-scale single weather image feature vector corresponding to each day.
4. The short term wind power prediction system for wind farms according to claim 3,
the first convolution unit is further used for; performing one-dimensional convolution coding on the one-dimensional feature vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first neighborhood scale single weather image feature vector;
wherein the formula is:
wherein a is the width of the first convolution kernel in the X direction, F is a first convolution kernel parameter vector, G is a local vector matrix operated with a convolution kernel function, w is the size of the first convolution kernel, and X represents a tension input vector;
the second convolution unit further configured to: performing one-dimensional convolution coding on the one-dimensional feature vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second neighborhood scale single weather image feature vector;
wherein the formula is:
wherein b is the width of the second convolution kernel in the X direction, F is a parameter vector of the second convolution kernel, G is a local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents a tension input vector.
5. The short-time wind power prediction system for a wind farm of claim 4, wherein the global meteorological data encoding module is further configured to: using each layer of the first convolutional neural network model as a feature extractor to respectively perform the following steps on input data in the forward direction transfer of the layer:
performing convolution processing on input data to obtain a convolution characteristic diagram;
performing mean pooling on the convolution feature map based on local channel dimensions to obtain a pooled feature map; and
performing nonlinear activation on the pooled feature map to obtain an activated feature map;
the output of the last layer of the first convolution neural network model as the feature extractor is the historical meteorological feature matrix, and the input of the first layer of the first convolution neural network model as the feature extractor is the two-dimensional feature matrix.
6. The short-term wind power prediction system for wind farms of claim 5, wherein the historical data feature fusion module is further configured to: fusing the historical meteorological feature matrix and the historical power feature matrix according to the following formula to obtain a fused feature matrix;
wherein the formula is:
M 1 =M*M 2
wherein M is 1 Representing said historical meteorological features matrix, M 2 Representing the historical power feature matrix, and M representing the fusion feature matrix.
7. The short-term wind power prediction system for a wind farm of claim 6, wherein the correction module is further configured to: based on the mean and variance of the feature value sets of all the positions of the decoding feature vector, correcting the feature values of all the positions in the decoding feature vector by the following formula to obtain the corrected decoding feature vector;
wherein the formula is:
wherein v is i Representing the feature values of the respective positions in the decoded feature vector, V representing the decoded feature vector, and μ and σ being feature sets V, respectively i E.g., the mean and variance of V, and L is the length of the decoded feature vector, and α is a weight hyperparameter.
8. The short-term wind power prediction system for a wind farm of claim 7, wherein the prediction value generation module is further configured to: decoding the corrected decoded feature vector using a plurality of fully-connected layers of the decoder to perform a decoding regression using the following formula to obtain the decoded value, wherein the formula is:wherein X is the corrected decoded feature vector, Y is the decoded value, W is a weight matrix, B is an offset vector,representing the matrix multiplication, h (-) is the activation function.
9. A short-time wind power prediction method for a wind power plant is characterized by comprising the following steps:
acquiring historical meteorological data and historical power data, wherein the meteorological data of each preset time point in the historical meteorological data comprise wind speed, wind direction, temperature, humidity and pressure;
arranging all data in the meteorological data of all preset time points in the historical meteorological data into input vectors, and then obtaining single-point multi-dimensional meteorological feature vectors through a full connection layer;
arranging the single-point multi-dimensional meteorological feature vectors into one-dimensional feature vectors by taking a single day as a time dimension, and then obtaining multi-scale single-meteorological feature vectors corresponding to each day through a multi-scale neighborhood feature extraction module;
arranging the multi-scale single weather image feature vectors corresponding to each day into a two-dimensional feature matrix, and then obtaining a historical weather feature matrix through a first convolution neural network model serving as a feature extractor;
arranging the power data of all preset time points in each day of the historical power data into power input vectors, and then obtaining single-day multi-scale power feature vectors corresponding to each day through the multi-scale neighborhood feature extraction module;
arranging the single-day multi-scale power characteristic vectors corresponding to each day into a two-dimensional characteristic matrix, and then obtaining a historical power characteristic matrix through a second convolutional neural network serving as a characteristic extractor;
fusing the historical meteorological feature matrix and the historical power feature matrix to obtain a fused feature matrix;
acquiring meteorological data of a plurality of preset time points of the day to be predicted and output power of a wind power plant to be predicted at the preset time points;
converting meteorological data of a plurality of preset time points on the day to be predicted into meteorological feature vectors to be predicted, converting output power of the wind power plant to be predicted at the preset time points into power feature vectors to be predicted, and fusing the meteorological feature vectors to be predicted and the power feature vectors to be predicted to obtain predicted feature vectors;
multiplying the predicted feature vector by the fusion feature matrix to obtain a decoded feature vector;
correcting the characteristic values of all positions in the decoding characteristic vector based on the mean value and the variance of the characteristic value sets of all the positions of the decoding characteristic vector to obtain a corrected decoding characteristic vector;
and performing decoding regression on the corrected decoding characteristic vector through a decoder to obtain a decoding value, wherein the decoding value is used for representing the power prediction value of the wind power plant to be predicted at the next preset time point.
10. The short-time wind power prediction method for the wind farm according to claim 9, wherein the step of arranging the items of data in the meteorological data of the historical meteorological data at the preset time points as input vectors and then obtaining single-point multi-dimensional meteorological feature vectors through a full connection layer comprises the following steps:
arranging various data in the meteorological data of each preset time point in the historical meteorological data into an input vector according to a time dimension; and
using the full-connection layer to perform full-connection coding on the input vector by using the following formula to extract high-dimensional implicit features of feature values of all positions in the input vector so as to obtain the single-point multi-dimensional meteorological feature vector, wherein the formula is as follows:wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the offset vector,representing a matrix multiplication.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211028966.7A CN115564092A (en) | 2022-08-26 | 2022-08-26 | Short-time wind power prediction system and method for wind power plant |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211028966.7A CN115564092A (en) | 2022-08-26 | 2022-08-26 | Short-time wind power prediction system and method for wind power plant |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115564092A true CN115564092A (en) | 2023-01-03 |
Family
ID=84739199
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211028966.7A Pending CN115564092A (en) | 2022-08-26 | 2022-08-26 | Short-time wind power prediction system and method for wind power plant |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115564092A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116362503A (en) * | 2023-03-30 | 2023-06-30 | 国网河南省电力公司安阳供电公司 | Electric power regulating method and system based on artificial intelligence |
-
2022
- 2022-08-26 CN CN202211028966.7A patent/CN115564092A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116362503A (en) * | 2023-03-30 | 2023-06-30 | 国网河南省电力公司安阳供电公司 | Electric power regulating method and system based on artificial intelligence |
CN116362503B (en) * | 2023-03-30 | 2023-11-07 | 国网河南省电力公司安阳供电公司 | Electric power regulating method and system based on artificial intelligence |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Khodayar et al. | Convolutional graph autoencoder: A generative deep neural network for probabilistic spatio-temporal solar irradiance forecasting | |
Cai et al. | Memory matching networks for one-shot image recognition | |
Ma et al. | A hybrid attention-based deep learning approach for wind power prediction | |
CN111832812A (en) | Wind power short-term prediction method based on deep learning | |
CN109471049B (en) | Satellite power supply system anomaly detection method based on improved stacked self-encoder | |
CN111967675A (en) | Photovoltaic power generation amount prediction method and prediction device | |
CN115577248A (en) | Fault diagnosis system and method for wind generating set | |
Li et al. | Deep spatio-temporal wind power forecasting | |
CN115951883B (en) | Service component management system of distributed micro-service architecture and method thereof | |
CN115099461A (en) | Solar radiation prediction method and system based on double-branch feature extraction | |
Ziyabari et al. | Multibranch attentive gated resnet for short-term spatio-temporal solar irradiance forecasting | |
CN115564092A (en) | Short-time wind power prediction system and method for wind power plant | |
CN116933922A (en) | WOA-CNN-LSTM-based photovoltaic power generation power prediction method | |
CN116404212B (en) | Capacity equalization control method and system for zinc-iron flow battery system | |
CN115357065B (en) | Remote intelligent dehumidification control system and method for offshore wind turbine | |
CN114793086B (en) | Double-sided double-glass photovoltaic panel adjusting bracket and adjusting method | |
CN115456259A (en) | Wind power plant site selection optimization system and method based on mesoscale data | |
CN116777039A (en) | Double-layer neural network wind speed prediction method based on training set segmentation and error correction | |
CN116124448A (en) | Fault diagnosis system and method for wind power gear box | |
CN116027874A (en) | Notebook computer power consumption control method and system thereof | |
CN115392525A (en) | Distributed wind power prediction method, device and equipment | |
CN114550159A (en) | Image subtitle generating method, device and equipment and readable storage medium | |
Ziyabari et al. | Multi-branch resnet-transformer for short-term spatio-temporal solar irradiance forecasting | |
Cai et al. | A Temporal Ensembling Based Semi-Supervised Graph Convolutional Network for Power Quality Disturbances Classification | |
CN117521606B (en) | Intelligent report generation system and method for financial data |
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 |