CN115330040A - Deep learning-based comprehensive energy distributed wind power generation prediction method and system - Google Patents
Deep learning-based comprehensive energy distributed wind power generation prediction method and system Download PDFInfo
- Publication number
- CN115330040A CN115330040A CN202210937499.3A CN202210937499A CN115330040A CN 115330040 A CN115330040 A CN 115330040A CN 202210937499 A CN202210937499 A CN 202210937499A CN 115330040 A CN115330040 A CN 115330040A
- Authority
- CN
- China
- Prior art keywords
- data
- power generation
- training
- prediction
- wind power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000010248 power generation Methods 0.000 title claims abstract description 67
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013135 deep learning Methods 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 56
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 claims abstract description 15
- 238000012545 processing Methods 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 16
- 238000007781 pre-processing Methods 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 10
- 230000014509 gene expression Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000013459 approach Methods 0.000 abstract description 2
- 238000005034 decoration Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006386 memory function Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000011160 research 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a comprehensive energy distributed wind power generation prediction method and system based on deep learning, which are combined with a CNN-GRU deep learning algorithm to accurately predict distributed wind power generation. Aiming at the missing values existing in the data source, the missing values are filled by using the weighted average value of the data points similar to the missing values based on the similarity principle, so that the problem of insufficient model training data caused by discarding is solved. According to the invention, through training the characteristic data transmitted by the CNN by using the GRU neural network, the time of model training is reduced while the relevance of the approach time is ensured.
Description
Technical Field
The invention belongs to the technical field of wind power generation and artificial intelligence, and particularly relates to a comprehensive energy distributed wind power generation prediction method and system based on deep learning.
Background
With the increasing energy crisis in recent years, the development and utilization of clean and sustainable energy are urgent, which is the focus of current research of relevant scholars. However, for the grid system, too many accesses of the distributed power generation devices will burden the grid management, and even cause grid accidents if the scheduling is not well performed. Therefore, at present, on one hand, the use of distributed clean energy needs to be increased as much as possible, and the use of traditional energy needs to be reduced, and on the other hand, for the prediction of distributed power generation data, it is very important to make a better scheduling strategy based on the prediction result. Distributed wind power generation is greatly affected by current meteorological data, such as: wind speed, wind direction and the like, and the randomness and the volatility of related meteorological data are large, so that the accurate prediction difficulty of distributed wind power generation is large at present. The prior art also has some problems while achieving certain prediction accuracy, such as:
(1) The current prediction method only considers the power generation condition of a single wind driven generator singly, does not start from the whole system, and ignores the correlation of the power generation data of each wind driven generator set.
(2) When the distributed wind power generation is predicted, the influence of meteorological data of a certain point on power generation data is always considered in isolation, and the correlation between the meteorological data at different moments is ignored.
Disclosure of Invention
In order to solve the problems, the invention discloses a comprehensive energy distributed wind power generation prediction method and a comprehensive energy distributed wind power generation prediction system based on deep learning.
In order to achieve the purpose, the technical scheme of the invention is as follows:
the comprehensive energy distributed wind power generation prediction method based on deep learning comprises the following steps:
step 1, preprocessing historical data
Acquiring historical data, dividing the historical data into a training set and a prediction set, and adding the historical wind power generation data of each wind generating set and meteorological data at corresponding moments into the two parts of data; the training set and the prediction set contain characteristic data of the input model, and corresponding labels are added into the training set; adding related weather forecast meteorological data into the forecast data; carrying out missing value filling and normalization processing on the data;
step 2, constructing a CNN neural network model
Constructing a convolution network model without a pooling layer, wherein the purpose is to extract a local dependency relationship among the generated energy of each wind generating set in a time dimension; the convolutional layer is composed of a plurality of filters having a width w and a height n, wherein the height is set to be the same as the number of variables; the kth filter scans the input matrix X and produces the following outputs:
h k ＝RELU(W K *X+b K ) (1)
wherein, is convolution operation, h k For the output vector, the relu function is relu = max (0,x); for h k Reducing the length thereof to T by filling 0; for the entire output, the network data structure of the convolutional layer is d c X T, wherein d c Representing the number of filters;
training a CNN network by adopting a training set, wherein output data is used for inputting into a GRU network;
step 3, constructing a GRU neural network training model
The GRU network comprises an updating gate and a resetting gate; wherein r is t 、z t The outputs of the reset gate and the refresh gate are X for the input at time t t ＝(x 1 ,x 1 ,…,x n ) Then the corresponding reset gate, update gate expressions are as follows:
r t ＝σ(w r ·[h t-1 ,x t ]+b r ) (2)
z t ＝σ(w z ·[h t-1 ,x t ]+b z ) (3)
updating the door z t : the method is mainly used for controlling the state information of the previous moment and the input information of the current moment, wherein the larger the value of the updating gate is, the more the state information of the previous moment is brought;
reset gate r t : like the refresh gate, the reset gate is used primarily to control how much previous state information is written to the candidate set
I.e., the larger the reset gate, the more state information is written at the previous time,
the following were used:
wherein
Representing the candidate hidden layer state of the GRU cell at the current moment, and controlling the output state of the GRU cell at the current moment;
receiving data output by the CNN network, and training the GRU network;
step 4, processing of predicted input characteristics
Constructing a data form which is the same as that of model training, wherein the input characteristics of the model are composed of two parts, namely historical power generation data and meteorological state data at a future moment, wherein the meteorological data at the future moment are obtained by combining weather forecast prediction data with historical similarity meteorological data; and if the data sampling frequency is 15min, the input characteristic corresponding to one wind generating set is represented by an expression (6), wherein p is the time span of historical wind power generation data, and q is the number of meteorological characteristic data:
X i ＝(x 1 ,x 2 ,…,x p ,x p+1 ,x p+2 ,…,x p+q ) (6)
weather data predicted for weather forecasts as in equation (7):
X q ＝(x p+1 ,x p+2 ,…,x p+q ) (7)
similarity calculation is carried out by using the formula (7) and historical meteorological data, the correlation is measured through a Pearson correlation coefficient, and the specific calculation mode is as shown in a formula (8):
by taking the first k historical date data with high similarity to the predicted day,
r x,y ＝[r 1 ,r 2 ,…,r 96 ] (9)
finally, acquiring meteorological data of the day to be predicted in a weighted average mode, as shown in a formula (9);
and merging the meteorological data of the day to be predicted into the historical wind power generation data to form a model input characteristic, and inputting the model input characteristic into the trained CNN and GRU models to obtain a wind power generation prediction result.
Further, the missing value filling and normalization processing in step 1 includes the following steps:
step 1.1, singular value processing is carried out on original data, and null value processing is carried out on data which do not accord with the normality;
step 1.2, for null data, selecting data t before the time, taking an average value, and performing similarity calculation with the previous historical data by using a similarity principle:
step 1.3, calculating missing value data by adopting a near-to-far weighted average mode for the acquired similar date data;
step 1.4, eliminating the influence of dimension in a normalization mode:
further, in step 1, the tags added in the training set are power generation amount data at corresponding moments, and the meteorological feature data in the training set and the prediction set include: the wind speed at the height of 10 meters, the wind direction at the height of 10 meters, the wind speed at the height of 30 meters, the wind direction at the height of 30 meters, the wind speed at the height of 50 meters, the wind direction at the height of 50 meters, the wind speed at the height of 70 meters, the wind direction at the height of 70 meters, the wind speed at the height of a fan hub, the wind direction at the height of the fan hub, the air temperature, the air pressure and the relative humidity.
The invention also provides a comprehensive energy distributed wind power generation prediction system based on deep learning, which comprises a data preprocessing module, a CNN module, a GRU module and a prediction module; the data preprocessing module is used for preprocessing data and generating a training set and a prediction set so as to realize the step 1 in the comprehensive energy distributed wind power generation prediction method based on deep learning; the CNN module is used for constructing and training a CNN neural network training model to realize the step 2 in the comprehensive energy distributed wind power generation prediction method based on deep learning; the GRU module is used for constructing and training a GRU neural network training model and realizing the step 3 in the comprehensive energy distributed wind power generation prediction method based on deep learning; and the prediction module is used for processing the prediction input characteristics, inputting the trained CNN and GRU models for prediction to obtain a prediction result, and realizing the step 4 in the comprehensive energy distributed wind power generation prediction method based on deep learning.
The invention has the beneficial effects that:
1. according to the method, the CNN-GRU deep learning algorithm is combined to accurately predict the distributed wind power generation, the CNN is used for extracting the correlation among the groups of data of each wind power generator, and then the correlation is input into the GRU, so that the GRU has a memory function of time sequence data, and the problem that the training speed is influenced by too much LSTM network weight is avoided, and the prediction timeliness and accuracy are guaranteed.
2. Aiming at the missing values existing in the data source, the missing values are filled by using the weighted average value of the data points similar to the missing values based on the similarity principle, so that the problem of insufficient model training data caused by discarding is solved.
3. The invention also aims to solve the technical problems in the prior art; the relevance of data in the time of approach of meteorological data and power generation data is fully considered, and therefore the effectiveness of the following GRU input features is guaranteed. By training the characteristic data transmitted by the CNN by using the GRU neural network, the time of model training is reduced to a certain extent while the near time correlation is ensured.
Drawings
FIG. 1 is a flow chart of a comprehensive energy distributed wind power generation prediction method based on deep learning provided by the invention.
Fig. 2 is a diagram of the GRU algorithm structure.
Detailed Description
The technical solutions provided by the present invention will be described in detail below with reference to specific examples, and it should be understood that the following specific embodiments are only illustrative of the present invention and are not intended to limit the scope of the present invention.
The invention provides a comprehensive energy distributed wind power generation prediction method based on deep learning, the flow of which is shown in figure 1, and the method comprises the following steps:
step 1, preprocessing historical data
The historical data is divided into a training set and a prediction set, because the number of the researched wind generating sets is large, the correlation among the wind generating sets is comprehensively considered, the historical wind power generation data of each wind generating set needs to be added into the two parts of data, and in addition, meteorological data at the corresponding moment also needs to be added; in addition, for the training set, a corresponding label, namely the power generation amount data at the corresponding moment, needs to be added. The training set and the prediction set contain feature data of the input model, the feature data contains corresponding feature data in addition to corresponding historical power generation data, and the meteorological feature data corresponding to the wind power generation data contains: wind speed at 10 m height, wind direction at 10 m height, wind speed at 30 m height, wind direction at 30 m height, wind speed at 50 m height, wind direction at 50 m height, wind speed at 70 m height, wind direction at 70 m height, wind speed at fan hub height, wind direction at fan hub height, air temperature, air pressure and relative humidity; for the training data, label data, i.e. wind power generation, is also included. The prediction data includes weather data related to weather prediction, in addition to the above-described feature data.
Data preprocessing is performed before the model is imported. The data preprocessing mainly comprises missing value filling and normalization processing of data; for a scene to which the method is applied, original data comprises partial missing data, the missing data is filled by utilizing a similarity principle, in addition, due to the fact that the characteristic data comprises meteorological data and related generated energy data, dimensions of all dimensions are inconsistent, prediction accuracy is affected, and therefore normalization processing needs to be carried out on the data before a model is imported. In addition, for the prediction data, the input features are selected by utilizing the similarity principle. The missing data mainly comprises two parts, namely data missing caused by sensor failure or artificial reasons; in addition, the data is singular values and is treated as missing values. The method comprises the steps of carrying out similarity calculation on the characteristic data near a missing value and historical data to obtain K points with the highest similarity, carrying out weighted average calculation to fill up the missing value, and obtaining the data with the highest similarity when K =7 according to experiments; for the predicted data, input feature data is calculated by using a similarity principle, weather data predicted by weather forecast is used, the similarity between the predicted weather data and historical weather data is calculated, the similarity is sorted according to absolute values, the first k data with high similarity are taken, and mean value processing is carried out on the data and the weather data predicted by the weather forecast, so that a weather data part of the feature data is obtained, and the highest correlation with original data is obtained when k =10 through experiments. And finally, normalizing the obtained data to eliminate the influence of the dimension on the model, so that the data characteristics are input to obtain the distributed wind power generation amount.
The specific calculation process of the data filling and data normalization processing comprises the following steps:
step 1.1, firstly, singular value processing is carried out on original data, and null value processing is carried out on data which do not conform to the normality;
step 1.2, for null data, by selecting data at t moment before the moment, averaging, and performing similarity calculation with the previous historical data by using a similarity principle, a similarity calculation formula can be calculated by using the following formula:
in the formula, n is the number of data, and x and y respectively represent two groups of meteorological data;
is the mean value of x, σ x Is the standard deviation of the measured data to be measured,
is the mean value of y, σ y Is the standard deviation.
Step 1.3, calculating missing value data by adopting a near-to-far weighted average mode for the acquired similar date data;
step 1.4, finally, in order to avoid the influence on the training error caused by the inconsistent dimension, the influence of the dimension is eliminated in a normalization mode;
wherein x is a characteristic value, x min Is the minimum value, x, of a set of characteristic values max Is the maximum value in a set of characteristic values.
Step 2, constructing a CNN neural network model: the CNN has a good effect on local feature extraction, and for a wind power generation system, the generated energy data among different wind generating sets has great correlation, so that the CNN model constructed here mainly aims at performing correlation extraction on features so as to better transmit the data into a GRU network. The CNN used by the invention is a convolution network without a pooling layer, and aims to extract the local dependence among the generated energy of each wind generating set in the time dimension. The convolutional layer is composed of a plurality of filters having a width w and a height n, wherein the height is set to be the same as the number of variables. The kth filter scans the input matrix X and produces the following outputs:
h k ＝RELU(W K *X+b K ) (1)
wherein, is convolution operation, h k For the output vector, the relu function is relu = max (0,x), W K As weights of the corresponding CNN neural network, b K Is the bias of the corresponding CNN neural network. For h k The length of which is reduced to T by filling 0. Thus for the entire output, the convolutional layer has a network data structure of d c X T, wherein d c The number of filters is indicated.
The CNN module mainly utilizes a CNN network to extract correlation characteristics of multiple wind generating sets, and mainly comprises two operation modes. Firstly, for training data, data input into a model comprise characteristic data and label data, and a CNN network can better fit the distribution of the data mainly by fitting the local correlation among all wind generating sets, so that a better prediction model is provided for distributed wind power generation; in addition, for prediction data, as the network structure parameters are determined in the training phase, the CNN network only needs to receive preprocessed feature data so as to fit the local correlation among the wind turbine group data, and finally the data flows into the GRU network.
Step 3, constructing a GRU neural network training model: GRU as a circulatory nerveThe network variant keeps the original circulating neural network with good memory function on time sequence data, avoids the problems of gradient disappearance and explosion to a certain extent, and compared with the LSTM neural network, avoids the excessive parameters generated by an input gate, an output gate and a forgetting gate, thereby accelerating the model training speed on the premise of ensuring the model performance. The GRU algorithm structure is shown in fig. 2, and for the GRU, only two gates are used, namely the refresh gate and the reset gate. Wherein r is t 、z t Outputs of reset gate and refresh gate, respectively, and input is X for time t t ＝(x 1 ,x 1 ,…,x n ) Then, the corresponding reset gate, update gate expressions are as follows:
r t ＝σ(w r ·[h t-1 ,x t ]+b r ) (2)
z t ＝σ(w z ·[h t-1 ,x t ]+b z ) (3)
where σ is the activation function, w r 、b r Weight and offset, w, of the reset gate, respectively z 、b z Respectively, the weight and bias of the update gate.
Updating the door z t : the method is mainly used for controlling the state information of the previous moment and the input information of the current moment, wherein the larger the value of the updating gate is, the more the state information of the previous moment is brought;
reset gate r t : like the refresh gate, the reset gate is used primarily to control how much previous state information is written to the candidate set
In the above, i.e., the larger the reset gate, the more state information of the previous time is written.
Wherein h is t Represents the unit state of the GRU cell at the current moment,
representing the candidate hidden layer state of GRU cell at the current time, for controlling the output state of GRU cell at the current time, h t-1 To a last timestamp state, w h 、b h Respectively the weight and offset of the memory, z t The door status is updated for the previous calculations.
The GRU module mainly utilizes a GRU algorithm to predict the wind power generation capacity and also comprises two operation modes. For training data, feature data and label data are imported, the feature data flow in from a CNN module, and the deviation between a predicted value and an actual label is calculated for multiple times, so that reverse correction of model parameters is performed, and finally a distributed wind power generation prediction model is obtained; for the predicted data, the GRU network receives the data from the CNN processing stream, and since the internal parameters are determined in the training stage, only the input features need to be combined for calculation, so as to obtain the final predicted result.
And 4, processing of prediction input characteristics: when distributed wind power generation prediction is carried out, the same data form as model training needs to be constructed; the input characteristics of the model are composed of two parts, namely historical power generation data and meteorological state data at the future moment, wherein the meteorological data at the future moment are obtained by combining weather forecast prediction data with historical similarity meteorological data. The data used by the method is sampling frequency once in 15min, and the input characteristic corresponding to one wind generating set can be expressed as an expression (6), wherein p is the time span of historical wind power generation data, and q is the number of meteorological characteristic data.
X i ＝(x 1 ,x 2 ,…,x p ,x p+1 ,x p+2 ,…,x p+q ) (6)
X i Is a set of features, x 1 …x p For historical wind power data, x p+1 …x p+q Is a meteorological data characteristic.
For meteorological feature data, historical similar data is added to serve as input features of model prediction, therefore, data used for calculation are mainly meteorological data and do not include historical power generation data, the formula (7) is meteorological data predicted by weather forecast, similarity calculation is conducted by using the formula (7) and the historical meteorological data, the correlation is measured through Pearson correlation coefficients, the specific calculation mode is the formula (8), historical date data with high similarity with the predicted date are obtained through the previous k historical date data, and finally the meteorological data of the date to be predicted are obtained through a weighted average mode, and the formula (9) is shown.
X q ＝(x p+1 ,x p+2 ,…,x p+q ) (7)
r x,y ＝[r 1 ,r 2 ,…,r 96 ] (9)
n is the number of historical data, and x and y respectively represent two groups of meteorological data;
is the mean value of x, σ x Is the standard deviation of the measured data to be measured,
is the average value of y, σ y Is the standard deviation.
And merging the meteorological data of the day to be predicted into the historical wind power generation data to form a model input characteristic, and inputting the model input characteristic into the trained CNN and GRU models to obtain a wind power generation prediction result.
The invention also provides a comprehensive energy distributed wind power generation prediction system based on deep learning, which comprises a data preprocessing module, a CNN module, a GRU module and a prediction module. The data preprocessing module is used for preprocessing data and generating a training set and a prediction set, and specifically realizes the step 1 in the comprehensive energy distributed wind power generation prediction method based on deep learning. And the CNN module is used for constructing and training a CNN neural network training model, and specifically realizing the step 2 in the comprehensive energy distributed wind power generation prediction method based on deep learning. The GRU module is used for constructing and training a GRU neural network training model, and particularly realizes the step 3 in the comprehensive energy distributed wind power generation prediction method based on deep learning. The prediction module is used for processing prediction input characteristics (the prediction data is preprocessed in the step 1), inputting the trained CNN and GRU models for prediction to obtain a prediction result, and specifically realizing the step 4 in the comprehensive energy distributed wind power generation prediction method based on deep learning.
It should be noted that the above-mentioned contents only illustrate the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and it is obvious to those skilled in the art that several modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations fall within the protection scope of the claims of the present invention.
Claims (4)
1. The comprehensive energy distributed wind power generation prediction method based on deep learning is characterized by comprising the following steps:
step 1, preprocessing historical data
Obtaining historical data, dividing the historical data into a training set and a prediction set, and adding the historical wind power generation data of each wind generating set and meteorological data at corresponding moments into the two parts of data; the training set and the prediction set contain characteristic data of the input model, and corresponding labels are added into the training set; adding related weather forecast meteorological data into the forecast data; missing value filling and normalization processing are carried out on the data;
step 2, constructing and training a CNN neural network model
Constructing a convolution network model without a pooling layer, wherein the purpose is to extract a local dependency relationship among the generated energy of each wind generating set in a time dimension; the convolutional layer is composed of a plurality of filters having a width w and a height n, wherein the height is set to be the same as the number of variables; the kth filter scans the input matrix X and produces the following outputs:
h k ＝RELU(W K *X+b K ) (1)
wherein, is convolution operation, h k For the output vector, the relu function is relu = max (0,x); for h k Reducing the length thereof to T by filling 0; for the entire output, the convolutional layer's network data structure is d c X T, wherein d c Representing the number of filters;
training a CNN network by adopting a training set, wherein output data is used for inputting into a GRU network;
step 3, building and training GRU neural network training model
The GRU network comprises an updating gate and a resetting gate; wherein r is t 、z t Outputs of reset gate and refresh gate, respectively, and input is X for time t t ＝(x 1 ,x 1 ,…,x n ) Then the corresponding reset gate, update gate expressions are as follows:
r t ＝σ(w r ·[h t-1 ,x t ]+b r ) (2)
z t ＝σ(w z ·[h t-1 ,x t ]+b z ) (3)
updating the door z t : the method is mainly used for controlling the state information of the previous moment and the input information of the current moment, wherein the larger the value of the updating gate is, the more the state information of the previous moment is brought;
reset gate r t : like the refresh gate, the reset gate is used primarily to control how much previous state information is written to the candidate set
I.e., the larger the reset gate, the more state information is written at the previous time,
the following:
wherein
Representing the candidate hidden layer state of the GRU cell at the current moment, and controlling the output state of the GRU cell at the current moment;
receiving data output by the CNN network, and training the GRU network;
step 4, processing of prediction input characteristics
The method comprises the following steps of constructing a data form which is the same as that of model training, wherein the input characteristics of a model are composed of two parts, namely historical power generation data and meteorological state data at a future moment, wherein the meteorological data at the future moment are obtained by combining weather forecast prediction data with historical similarity meteorological data; and if the data sampling frequency is 15min, the input characteristic corresponding to one wind generating set is represented by an expression (6), wherein p is the time span of historical wind power generation data, and q is the number of meteorological characteristic data:
X i ＝(x 1 ,x 2 ,…,x p ,x p+1 ,x p+2 ,…,x p+q ) (6)
weather data predicted for weather forecasts as in equation (7):
X q ＝(x p+1 ,x p+2 ,…,x p+q ) (7)
similarity calculation is carried out by using the formula (7) and historical meteorological data, the correlation is measured through a Pearson correlation coefficient, and the specific calculation mode is as shown in a formula (8):
by taking the first k historical date data with high similarity to the predicted day,
r x,y ＝[r 1 ,r 2 ,…,r 96 ] (9)
finally, acquiring meteorological data of the day to be predicted in a weighted average mode, wherein the meteorological data are shown as a formula (9);
and merging the meteorological data of the day to be predicted into the historical wind power generation data to form a model input characteristic, and inputting the model input characteristic into the trained CNN and GRU models to obtain a wind power generation prediction result.
2. The deep learning-based integrated energy distributed wind power generation prediction method according to claim 1, wherein the missing value filling and normalization process in step 1 comprises the following steps:
step 1.1, singular value processing is carried out on original data, and null value processing is carried out on data which do not accord with the normative principle;
step 1.2, for null data, selecting data at t moment before the moment, taking an average value, and performing similarity calculation with the previous historical data by using a similarity principle:
step 1.3, calculating missing value data by adopting a near-to-far weighted average mode for the acquired similar date data;
step 1.4, eliminating the influence of dimension in a normalization mode:
3. the deep learning-based comprehensive energy distributed wind power generation prediction method according to claim 1, wherein in the step 1, the labels added into the training set are power generation amount data at corresponding moments, and the meteorological feature data in the training set and the prediction set comprise: the wind speed at the height of 10 meters, the wind direction at the height of 10 meters, the wind speed at the height of 30 meters, the wind direction at the height of 30 meters, the wind speed at the height of 50 meters, the wind direction at the height of 50 meters, the wind speed at the height of 70 meters, the wind direction at the height of 70 meters, the wind speed at the height of a fan hub, the wind direction at the height of the fan hub, the air temperature, the air pressure and the relative humidity.
4. The comprehensive energy distributed wind power generation prediction system based on deep learning comprises a data preprocessing module, a CNN module, a GRU module and a prediction module; the method is characterized in that the data preprocessing module is used for preprocessing data and generating a training set and a prediction set, so as to realize the step 1 in the deep learning-based comprehensive energy distributed wind power generation prediction method of any one of the preceding claims 1 to 3; the CNN module is used for constructing and training a CNN neural network training model to realize the step 2 in the deep learning-based comprehensive energy distributed wind power generation prediction method of any one of the preceding claims 1-3; the GRU module is used for constructing and training a GRU neural network training model and realizing the step 3 in the comprehensive energy distributed wind power generation prediction method based on deep learning; the prediction module is used for processing the prediction input characteristics, inputting the trained CNN and GRU models for prediction to obtain a prediction result, and realizing the step 4 in the deep learning-based comprehensive energy distributed wind power generation prediction method of any one of the claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210937499.3A CN115330040A (en) | 2022-08-05 | 2022-08-05 | Deep learning-based comprehensive energy distributed wind power generation prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210937499.3A CN115330040A (en) | 2022-08-05 | 2022-08-05 | Deep learning-based comprehensive energy distributed wind power generation prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115330040A true CN115330040A (en) | 2022-11-11 |
Family
ID=83920969
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210937499.3A Pending CN115330040A (en) | 2022-08-05 | 2022-08-05 | Deep learning-based comprehensive energy distributed wind power generation prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115330040A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116341644A (en) * | 2023-03-30 | 2023-06-27 | 山东大学 | Modeling method and system for physical guidance and data driving integrated energy system equipment |
| CN117875506A (en) * | 2024-01-16 | 2024-04-12 | 盐城工学院 | Method for predicting and processing aquaculture tail water based on LSTM neural network model |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108448610A (en) * | 2018-03-12 | 2018-08-24 | 华南理工大学 | A kind of short-term wind power prediction method based on deep learning |
| CN111832812A (en) * | 2020-06-27 | 2020-10-27 | 南通大学 | Wind power short-term prediction method based on deep learning |
| CN112365093A (en) * | 2020-12-01 | 2021-02-12 | 国家海洋信息中心 | GRU deep learning-based multi-feature factor red tide prediction model |
| CN112949931A (en) * | 2021-03-19 | 2021-06-11 | 北京交通大学 | Method and device for predicting charging station data with hybrid data drive and model |
| CN113191091A (en) * | 2021-06-03 | 2021-07-30 | 上海交通大学 | Wind speed prediction method, system and equipment based on hybrid deep learning mechanism |
| CN113822482A (en) * | 2021-09-24 | 2021-12-21 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Method and device for establishing load prediction model of comprehensive energy system |
| CN113837499A (en) * | 2021-11-24 | 2021-12-24 | 中国电建集团江西省电力设计院有限公司 | Ultra-short-term wind power prediction method and system |
-
2022
- 2022-08-05 CN CN202210937499.3A patent/CN115330040A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108448610A (en) * | 2018-03-12 | 2018-08-24 | 华南理工大学 | A kind of short-term wind power prediction method based on deep learning |
| CN111832812A (en) * | 2020-06-27 | 2020-10-27 | 南通大学 | Wind power short-term prediction method based on deep learning |
| CN112365093A (en) * | 2020-12-01 | 2021-02-12 | 国家海洋信息中心 | GRU deep learning-based multi-feature factor red tide prediction model |
| CN112949931A (en) * | 2021-03-19 | 2021-06-11 | 北京交通大学 | Method and device for predicting charging station data with hybrid data drive and model |
| CN113191091A (en) * | 2021-06-03 | 2021-07-30 | 上海交通大学 | Wind speed prediction method, system and equipment based on hybrid deep learning mechanism |
| CN113822482A (en) * | 2021-09-24 | 2021-12-21 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Method and device for establishing load prediction model of comprehensive energy system |
| CN113837499A (en) * | 2021-11-24 | 2021-12-24 | 中国电建集团江西省电力设计院有限公司 | Ultra-short-term wind power prediction method and system |
Non-Patent Citations (1)
| Title |
|---|
| 高野: "社区综合能源***供需预测技术研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》, no. 3, 15 March 2022 (2022-03-15), pages 039 - 97 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116341644A (en) * | 2023-03-30 | 2023-06-27 | 山东大学 | Modeling method and system for physical guidance and data driving integrated energy system equipment |
| CN117875506A (en) * | 2024-01-16 | 2024-04-12 | 盐城工学院 | Method for predicting and processing aquaculture tail water based on LSTM neural network model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| An et al. | Short-term wind power prediction based on particle swarm optimization-extreme learning machine model combined with AdaBoost algorithm | |
| Shi et al. | Hybrid forecasting model for very-short term wind power forecasting based on grey relational analysis and wind speed distribution features | |
| CN103390116B (en) | Use the photovoltaic power station power generation power forecasting method of stepping mode | |
| CN115330040A (en) | Deep learning-based comprehensive energy distributed wind power generation prediction method and system | |
| CN109802430A (en) | A kind of wind-powered electricity generation power grid control method based on LSTM-Attention network | |
| CN102479339B (en) | Method and system for forecasting short-term wind speed of wind farm based on hybrid neural network | |
| Mabel et al. | Estimation of energy yield from wind farms using artificial neural networks | |
| CN105046374A (en) | Power interval predication method based on nucleus limit learning machine model | |
| CN112215428B (en) | Photovoltaic power generation power prediction method and system based on error correction and fuzzy logic | |
| CN109255726A (en) | A kind of ultra-short term wind power prediction method of Hybrid Intelligent Technology | |
| CN102479347A (en) | Method and system for forecasting short-term wind speed of wind farm based on data driving | |
| CN107204615A (en) | A kind of method and system for realizing power prediction | |
| Li et al. | Short-term prediction of the power of a new wind turbine based on IAO-LSTM | |
| CN111404204A (en) | Optimized scheduling method for power system including wind power plant by using improved wolf pack algorithm | |
| CN114757427A (en) | Autoregressive corrected LSTM intelligent wind power plant ultra-short term power prediction method | |
| CN113962441A (en) | Short-term irradiance prediction method and prediction system based on historical data analysis | |
| CN112072643A (en) | Light-storage system online scheduling method based on depth certainty gradient strategy | |
| CN105741192B (en) | Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant | |
| CN107358059A (en) | Short-term photovoltaic energy Forecasting Methodology and device | |
| Ranganayaki et al. | Svm based neuro fuzzy model for short term wind power forecasting | |
| Yang et al. | Day‐ahead wind power combination forecasting based on corrected numerical weather prediction and entropy method | |
| CN116128211A (en) | Wind-light-water combined short-term optimization scheduling method based on wind-light uncertainty prediction scene | |
| CN105787813A (en) | Short-term wind power prediction method based on covariance | |
| Yuan et al. | Multi-area generation expansion planning model of high variable generation penetration | |
| Wan et al. | Turbine location wind speed forecast using convolutional neural network |
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 |