CN114298130A - Short-term wind power prediction method and device - Google Patents

Short-term wind power prediction method and device Download PDF

Info

Publication number
CN114298130A
CN114298130A CN202111334208.3A CN202111334208A CN114298130A CN 114298130 A CN114298130 A CN 114298130A CN 202111334208 A CN202111334208 A CN 202111334208A CN 114298130 A CN114298130 A CN 114298130A
Authority
CN
China
Prior art keywords
wind power
data
point
characteristic
sample
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
Application number
CN202111334208.3A
Other languages
Chinese (zh)
Inventor
叶林
王森
林昇
张伟
王爽
蒋贲
王建国
陈志文
刘庭
杜洋
张小贝
冯翔宇
陈卓
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaneng Renewables Corp Ltd
Beijing Huaneng Xinrui Control Technology Co Ltd
Original Assignee
Huaneng Renewables Corp Ltd
Beijing Huaneng Xinrui Control Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huaneng Renewables Corp Ltd, Beijing Huaneng Xinrui Control Technology Co Ltd filed Critical Huaneng Renewables Corp Ltd
Priority to CN202111334208.3A priority Critical patent/CN114298130A/en
Publication of CN114298130A publication Critical patent/CN114298130A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a short-term wind power prediction method and a short-term wind power prediction device, wherein the method comprises the following steps: acquiring fan operation data of a wind power plant; removing the abnormal value of the wind power in the operation data; selecting characteristic variables influencing wind power; establishing a wind power prediction model by adopting a deep neural network based on the characteristic variables; and evaluating the prediction result of the wind power prediction model. Abnormal data are screened and removed by adopting an improved fuzzy C-means clustering algorithm, and a reliable data basis is provided for the prediction modeling of the wind power. And selecting a characteristic variable influencing wind power as a characteristic input through a Relieff algorithm. The Bi-LSTM neural network algorithm is adopted for modeling and predicting the wind power, and compared with the traditional BP neural network, RNN, LSTM and other algorithms, the method has higher prediction precision and realizes accurate prediction of short-term wind power.

Description

Short-term wind power prediction method and device
Technical Field
The disclosure belongs to the technical field of wind power prediction, and particularly relates to a short-term wind power prediction method and device.
Background
The prediction for wind power is mainly based on statistical models and machine learning models. Some of the most common methods in statistical models are Autoregressive (AR), Moving Average (MA), autoregressive integrated moving average model (ARMA), autoregressive moving average model (ARMA), and the use of kalman filters. These methods use classical time series analysis to model statistical relationships between historical data. This approach relies on learning from historical data. A large amount of data was analyzed. The model is usually represented as a time series model or dynamic model, the parameters of which are estimated by minimizing a cost function over the historical training data set. Machine learning models have been widely used in recent years, and prediction is generally performed using linear and nonlinear models of Artificial Neural Networks (ANN), markov chains, fuzzy systems, Support Vector Machines (SVM), Quantile Regression (QR), k-nearest neighbors (KNN), Extreme Learning Machines (ELM), Random Forests (RF), multi-layer perceptrons (MLP), and dynamic bayesian clustering and support vector regression. Another nonlinear predictive model Gaussian Process (GP) is proposed for wind power prediction. These machine learning models require a large amount of data to learn the relationships between variables and are weak in their ability to extract features within the data.
Disclosure of Invention
The present disclosure is directed to at least one of the technical problems in the prior art, and provides a short-term wind power prediction method and apparatus.
In one aspect of the present disclosure, a short-term wind power prediction method is provided, where the method includes:
acquiring fan operation data of a wind power plant;
removing the abnormal value of the wind power in the operation data;
selecting characteristic variables influencing wind power;
establishing a wind power prediction model by adopting a deep neural network based on the characteristic variables;
and evaluating the prediction result of the wind power prediction model.
In some embodiments, the removing the wind power abnormal value in the operation data includes:
and eliminating the wind power abnormal value in the operating data by adopting an improved fuzzy C-means clustering algorithm.
In some embodiments, the removing wind power abnormal values in the operating data by using an improved fuzzy C-means clustering algorithm includes:
clustering the wind power data to obtain a cluster number i and a cluster center xk(ii) a Calculating the mean square error corresponding to the clustering center at each moment:
Figure BDA0003349921240000021
xipoints in the sample set, n is the number of samples;
judging abnormal data in the data by the following relational expression (2), and if the following relational expression (2) is satisfied, determining the abnormal data as the abnormal data:
Figure BDA0003349921240000022
the abnormal data is corrected by using the following relational expression (3):
Figure BDA0003349921240000023
where p and q are abnormal data points, XdFor the curve to be examined, XtIs a characteristic curve, XrIs a correction curve; xr(i) To correct the ith point of the curve, Xt(i) Is the i-th point of the characteristic curve, Xd(p-1) is the p-1 st point of the examined curve, Xd(q +1) is the q +1 th point of the detected curve, Xt (p-1) is the p-1 th point of the characteristic curve, and Xt (q +1) is the q +1 th point of the characteristic curve.
In some embodiments, the clustering is performed on the wind power data to obtain a cluster number i and a cluster center xkThe method comprises the following steps:
obtaining the upper limit of the clustering number through a classification algorithm, and determining the number of clustering centers in advance by applying fuzzy C-means clustering on the basis to ensure algorithm convergence so as to obtain a global optimal solution;
fuzzy C-means clustering algorithm:
the objective function of the fuzzy C-means clustering is:
Figure BDA0003349921240000031
m>1 is a blurring factor, μij(i-1, 2 …, M; j-1, 2, …, R) represents the degree of membership of the ith sample in the jth class, | xi-vj |2Denotes the distance, V, from the ith sample to the jth cluster centerjIs a clustering center;
the updating of the iterative center and membership is performed by the following relations (5) and (6):
Figure BDA0003349921240000032
Figure BDA0003349921240000033
r represents iteration times, whether | J (r +1) -J (r) | < epsilon is checked, if yes, calculation is stopped, and if not, iteration is continued;
and (3) a subtraction and classification algorithm:
for each point X in the sample set XiThe density value index is calculated according to the following relations (7) and (8):
Figure BDA0003349921240000034
Figure BDA0003349921240000035
selecting the data point x with the highest density indexc1(Density index is D)c1) As a first cluster center; gamma rayaA point-to-point x representing the radius of the area at that point, outside the radiusiContribution of density index ofAre small; k-th selected clustering center xck(Density index is D)ck) The density index of each point is corrected by the following relational expression (9):
Figure BDA0003349921240000041
γbis a positive number, typically taken as γb＝1.5γaRepresenting a region in which the density indicator function is significantly reduced;
selecting the point x with the highest density index from the corrected data pointsck+1As a new cluster center; whether the following relation (10) is satisfied or not is judged, and if not, the relation (5) and the relation (6) are returned until the condition is satisfied.
Figure BDA0003349921240000042
In some embodiments, the selecting the characteristic variables affecting the wind power includes:
and selecting characteristic variables influencing wind power by adopting a characteristic extraction Relieff algorithm.
In some embodiments, the selecting a characteristic variable affecting wind power by using a characteristic extraction ReliefF algorithm includes:
selecting a sample x from the training samplesiSelect k xiThe nearest neighbor sample of (1) is recorded as H;
from xiSelecting k different classes of non-similar nearest neighbor samples to be recorded as M (c);
obtaining a characteristic weight by calculating the intra-class distance and the inter-class distance of the nearest neighbor sample;
by repeating the above process, the weights of all the features are finally obtained.
In some embodiments, the formula for the ReliefF algorithm to update the feature weights is:
Figure BDA0003349921240000043
wherein A is0A feature set that is an original data set; a represents a feature subset of the filtered data set; w [ A ]0]The weight coefficient before updating; w [ A ]]Is the updated weight coefficient; x is the number ofiIs the ith sample, H represents the sum of xiNearest neighbor samples in the same class; diff (A, x)iH) is each feature x in AiAnd H, the calculation formula is as follows:
Figure BDA0003349921240000051
k is the number of nearest neighbors; p (C) is the ratio of the target sample C to the total sample; p (class (x)i) Is x is containediThe ratio of the same type of sample to the total sample; mj (C) represents the jth neighbor sample that is not of the same class as the target sample C; diff (A, x)iMj (C) is each feature x in AiAnd quantitative characterization of the differences between Mj (C).
In some embodiments, the building a wind power prediction model by using a deep neural network based on the characteristic variables includes:
and constructing the wind power prediction model by adopting a Bi-directional long-short term memory neural network Bi-LSTM based on the characteristic variables.
In some embodiments, the evaluating the prediction result of the wind power prediction model includes:
and evaluating the prediction result of the wind power prediction model through the average absolute error and the root mean square error.
In another aspect of the present disclosure, there is provided a short-term wind power prediction apparatus, the apparatus including:
the acquisition module is used for acquiring fan operation data of the wind power plant;
the removing module is used for removing the wind power abnormal value in the operation data;
the selection module is used for selecting characteristic variables influencing wind power;
the establishing module is used for establishing a wind power prediction model by adopting a deep neural network based on the characteristic variables;
and the evaluation module is used for evaluating the prediction result of the wind power prediction model.
According to the short-term wind power prediction method and device, the abnormal data are screened and removed by adopting the improved fuzzy C-means clustering algorithm, and a reliable data basis is provided for the prediction modeling of the wind power. And selecting a characteristic variable influencing wind power as a characteristic input through a Relieff algorithm. The Bi-LSTM neural network algorithm is adopted for modeling and predicting the wind power, and compared with the traditional BP neural network, RNN, LSTM and other algorithms, the method has higher prediction precision and realizes accurate prediction of short-term wind power.
Drawings
Fig. 1 is a flowchart of a short-term wind power prediction method according to an embodiment of the present disclosure;
FIG. 2 is a feature variable weight graph according to another embodiment of the present disclosure;
FIG. 3 is a Bi-LSTM neural network modeling schematic diagram of another embodiment of the present disclosure;
FIG. 4 is a comparison graph of predicted results for another embodiment of the present disclosure;
fig. 5 is a schematic structural diagram of a short-term wind power prediction apparatus according to another embodiment of the present disclosure.
Detailed Description
For a better understanding of the technical aspects of the present disclosure, reference is made to the following detailed description taken in conjunction with the accompanying drawings.
One aspect of the present embodiment, as shown in fig. 1, relates to a short-term wind power prediction method S100, where the method S100 includes:
and S110, acquiring fan operation data of the wind power plant.
Specifically, in this step, the wind turbine operation data includes wind speed, power, environmental factors, and the like of the wind turbine in the wind farm. Specifically actual measurement operation data of wind speed and power of a fan in the wind power plant and meteorological factor data of a corresponding date.
And S120, eliminating the wind power abnormal value in the operation data.
Specifically, in this step, an improved fuzzy C-means clustering algorithm (FCM) may be used for the screening of abnormal data.
As an example, the specific steps of screening abnormal data by using the improved fuzzy C-means clustering algorithm (FCM) include:
clustering the wind power data to obtain a cluster number i and a cluster center xk(ii) a Calculating the mean square error corresponding to the clustering center at each moment:
Figure BDA0003349921240000061
wherein x isiIs the point in the sample set and n is the number of samples.
Judging abnormal data in the data by the following relational expression (2), and if the following relational expression (2) is satisfied, determining the abnormal data as the abnormal data:
Figure BDA0003349921240000071
the abnormal data is corrected by using the following relational expression (3):
Figure BDA0003349921240000072
where p and q are abnormal data points, XdFor the curve to be examined, XtIs a characteristic curve, XrIs a correction curve; xr(i) To correct the ith point of the curve, Xt(i) Is the i-th point of the characteristic curve, Xd(p-1) is the p-1 st point of the examined curve, Xd(q +1) is the q +1 point of the examined curve, Xt(p-1) is the p-1 st point of the characteristic curve, Xt(q +1) is the q +1 th point of the characteristic curve.
The cluster number and the cluster center can be obtained by adopting the following modes:
clustering is carried out on the wind power data to obtain a clustering number i and a clustering center xkThe method comprises the following steps:
obtaining the upper limit of the clustering number through a classification algorithm, and determining the number of clustering centers in advance by applying fuzzy C-means clustering on the basis to ensure algorithm convergence so as to obtain a global optimal solution;
fuzzy C-means clustering algorithm:
the objective function of the fuzzy C-means clustering is:
Figure BDA0003349921240000073
m>1 is a blurring factor, μij(i-1, 2 …, M; j-1, 2, …, R) represents the degree of membership of the ith sample in the jth class, | xi-vj |2Denotes the distance, V, from the ith sample to the jth cluster centerjIs the cluster center.
The updating of the iterative center and membership is performed by the following relations (5) and (6):
Figure BDA0003349921240000081
Figure BDA0003349921240000082
r represents the iteration number, whether | J (r +1) -J (r) | < epsilon is checked, if yes, the calculation is stopped, and if not, the iteration is continued.
And (3) a subtraction and classification algorithm:
for each point X in the sample set XiThe density value index is calculated according to the following relations (7) and (8):
Figure BDA0003349921240000083
Figure BDA0003349921240000084
selecting the data point x with the highest density indexc1(Density index is D)c1) As a first cluster center; gamma rayaA point-to-point x representing the radius of the area at that point, outside the radiusiThe density index contribution of (a) is small; k-th selected clustering center xck(Density index is D)ck) The density index of each point is corrected by the following relational expression (9):
Figure BDA0003349921240000085
selecting the point x with the highest density index from the corrected data pointsck+1As a new cluster center; whether the following relation (10) is satisfied or not is judged, and if not, the relation (5) and the relation (6) are returned until the condition is satisfied.
Figure BDA0003349921240000086
And S130, selecting characteristic variables influencing wind power.
Specifically, in this step, a feature extraction ReliefF algorithm may be adopted to select a feature variable that affects the wind power. The Relieff algorithm is a widely applied filter-based feature selection model and has high classification efficiency. The algorithm is not limited by the type of data and therefore can effectively handle nominal or continuous features, data loss, noise tolerance, and the like. The principle of the algorithm is to assign weights to features based on class relevance and to select an optimal subset of features based on the weights. Thus, the stronger the relevance of the classification, the closer the similar samples are.
In some embodiments, the selecting a characteristic variable affecting wind power by using a characteristic extraction ReliefF algorithm includes:
selecting from training samplesTaking a sample xiSelect k xiThe nearest neighbor sample of (1) is recorded as H;
from xiSelecting k different classes of non-similar nearest neighbor samples to be recorded as M (c);
obtaining a characteristic weight by calculating the intra-class distance and the inter-class distance of the nearest neighbor sample;
by repeating the above process, the weights of all the features are finally obtained.
In some embodiments, the formula for the ReliefF algorithm to update the feature weights is:
Figure BDA0003349921240000091
wherein A is0A feature set that is an original data set; a represents a feature subset of the filtered data set; w [ A ]0]The weight coefficient before updating; w [ A ]]Is the updated weight coefficient; x is the number ofiIs the ith sample, H represents the sum of xiNearest neighbor samples in the same class; diff (A, x)iH) is each feature x in AiAnd H, the calculation formula is as follows:
Figure BDA0003349921240000092
k is the number of nearest neighbors; p (C) is the ratio of the target sample C to the total sample; p (class (x)i) Is x is containediThe ratio of the same type of sample to the total sample; mj (C) represents the jth neighbor sample that is not of the same class as the target sample C; diff (A, x)iMj (C) is each feature x in AiAnd quantitative characterization of the differences between Mj (C). The characteristic weights for each variable are derived as shown in fig. 2. Calculating the characteristic weight of each characteristic variable relative to the wind power is shown in table 1 below:
TABLE 1 characteristic weight of each characteristic variable with respect to wind power
Figure BDA0003349921240000101
And S140, establishing a wind power prediction model by adopting a deep neural network based on the characteristic variables.
Specifically, in this step, the wind power prediction model may be constructed by using a Bi-directional long-short term memory neural network Bi-LSTM based on the characteristic variables.
The recurrent neural network RNN has strong data processing capacity for data with strong correlation before and after time, and can realize short-term prediction of wind power. The LSTM model can avoid the problems of RNN gradient disappearance and gradient explosion, the long-term and short-term memory network can well notice the correlation before and after data, the gating cycle unit can remember the previous data and can selectively forget the data, so that the prediction is more accurate and faster, and the method is suitable for ultra-short-term wind power prediction. The Bi-LSTM is formed by combining a forward LSTM and a backward LSTM, the forward LSTM can acquire past data information of an input sequence, the backward LSTM can acquire future data information of the input sequence, and the forward LSTM and the backward LSTM train the time sequence twice. The method finally selects the Bi-directional long-short term memory neural network Bi-LSTM to construct a wind power prediction model, and compares different algorithms.
Further, after characteristic variables influencing wind power are selected as characteristic inputs, LSTM, Bi-LSTM, RNN and BP neural networks are respectively adopted for modeling, short-term wind power is predicted, and the parameters of each algorithm model are shown in the following table 2:
TABLE 2 Algorithm model parameter Table
Figure BDA0003349921240000102
Figure BDA0003349921240000111
The prediction is carried out for 24 hours of wind power a day, wherein the Bi-LSTM neural network modeling flow is shown in figure 3. The prediction results obtained by the different models are shown in fig. 4: the change trend of the wind power can be well tracked by the four models, wherein the LSTM and the Bi-LSTM have better effects, and compared with the LSTM, the Bi-LSTM can accurately predict the wind power in places with larger wind power fluctuation and shows good robustness.
S150, evaluating the prediction result of the wind power prediction model.
Specifically, in this step, a model evaluation index is introduced for quantitative analysis of the prediction result. The average absolute error (MAE) and the Root Mean Square Error (RMSE) are used as the evaluation indexes of the model accuracy in the prediction result,
Figure BDA0003349921240000112
Figure BDA0003349921240000113
in the formula (I), the compound is shown in the specification,
Figure BDA0003349921240000114
indicates the predicted value of power, piThe actual value of the power and N is the number of predicted samples. The evaluation indices of the different models are shown in table 3 below:
TABLE 3 evaluation indices of different models
Figure BDA0003349921240000115
Figure BDA0003349921240000121
As can be seen from the data in the table, the other three models achieve better prediction effects compared with the BP neural network, and compared with RNN, LSTM and Bi-LSTM both show more excellent prediction performance, wherein the mean absolute error and the root mean square error of Bi-LSTM are 4.1250 and 4.5871 respectively. The method has the advantages that the relation among variables can be better learned, the prediction precision is greatly improved through forward and backward LSTM training, a good prediction effect is obtained, and the method meets the requirement of short-term wind power prediction of a wind power plant.
According to the short-term wind power prediction method, abnormal data are screened and removed by adopting an improved fuzzy C-means clustering algorithm, and a reliable data basis is provided for the prediction modeling of the wind power. And selecting a characteristic variable influencing wind power as a characteristic input through a Relieff algorithm. The Bi-LSTM neural network algorithm is adopted for modeling and predicting the wind power, and compared with the traditional BP neural network, RNN, LSTM and other algorithms, the method has higher prediction precision and realizes accurate prediction of short-term wind power.
In summary, the short-term wind power prediction method disclosed by the invention is used for carrying out screening and elimination of abnormal values based on an improved fuzzy C-means clustering algorithm aiming at the short-term wind power prediction of a wind power plant, selecting characteristic variables as characteristic inputs through a Relieff algorithm aiming at the influence factors of the wind power, and carrying out modeling prediction on the wind power by adopting a Bi-LSTM neural network. The traditional BP neural network, RNN, LSTM and Bi-LSTM algorithms are respectively applied to modeling prediction, and the results show that the Bi-LSTM can more accurately track the change trend of the wind power and realize short-term accurate prediction of the wind power plant.
In another aspect of the present disclosure, as shown in fig. 5, there is provided a short-term wind power prediction apparatus 100, the apparatus 100 being adapted to the method described above, the apparatus 100 comprising:
the obtaining module 110 is configured to obtain wind turbine operation data of the wind farm.
And the eliminating module 120 is configured to eliminate the wind power abnormal value in the operating data.
And the selecting module 130 is used for selecting characteristic variables influencing the wind power.
And the establishing module 140 is configured to establish a wind power prediction model by using a deep neural network based on the characteristic variables.
And the evaluation module 150 is used for evaluating the prediction result of the wind power prediction model.
The short-term wind power prediction device of the embodiment screens and eliminates abnormal data by adopting an improved fuzzy C-means clustering algorithm, and provides a reliable data basis for wind power prediction modeling. And selecting a characteristic variable influencing wind power as a characteristic input through a Relieff algorithm. The Bi-LSTM neural network algorithm is adopted for modeling and predicting the wind power, and compared with the traditional BP neural network, RNN, LSTM and other algorithms, the method has higher prediction precision and realizes accurate prediction of short-term wind power.
It is to be understood that the above embodiments are merely exemplary embodiments that are employed to illustrate the principles of the present disclosure, and that the present disclosure is not limited thereto. It will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the disclosure, and these are to be considered as the scope of the disclosure.

Claims (10)

1. A short-term wind power prediction method is characterized by comprising the following steps:
acquiring fan operation data of a wind power plant;
removing the abnormal value of the wind power in the operation data;
selecting characteristic variables influencing wind power;
establishing a wind power prediction model by adopting a deep neural network based on the characteristic variables;
and evaluating the prediction result of the wind power prediction model.
2. The method according to claim 1, wherein the removing the wind power abnormal value in the operation data comprises:
and eliminating the wind power abnormal value in the operating data by adopting an improved fuzzy C-means clustering algorithm.
3. The method according to claim 2, wherein the removing the wind power abnormal value in the operation data by using the improved fuzzy C-means clustering algorithm comprises:
clustering the wind power data to obtain a cluster number i and a cluster center xk(ii) a Calculating the mean square error corresponding to the clustering center at each moment:
Figure FDA0003349921230000011
xipoints in the sample set, n is the number of samples;
judging abnormal data in the data by the following relational expression (2), and if the following relational expression (2) is satisfied, determining the abnormal data as the abnormal data:
Figure FDA0003349921230000012
the abnormal data is corrected by using the following relational expression (3):
Figure FDA0003349921230000013
where p and q are abnormal data points, XdFor the curve to be examined, XtIs a characteristic curve, XrIs a correction curve; xr(i) To correct the ith point of the curve, Xt(i) Is the i-th point of the characteristic curve, Xd(p-1) is the p-1 st point of the examined curve, Xd(q +1) is the q +1 point of the examined curve, Xt(p-1) is the p-1 st point of the characteristic curve, Xt(q +1) is the q +1 th point of the characteristic curve.
4. The method of claim 3, wherein the clustering of the wind power data results in a cluster number i and a cluster center xkThe method comprises the following steps:
obtaining the upper limit of the clustering number through a classification algorithm, and determining the number of clustering centers in advance by applying fuzzy C-means clustering on the basis to ensure algorithm convergence so as to obtain a global optimal solution;
fuzzy C-means clustering algorithm:
the objective function of the fuzzy C-means clustering is:
Figure FDA0003349921230000021
m>1 is a blurring factor, μij(i-1, 2 …, M; j-1, 2, …, R) represents the degree of membership of the ith sample in the jth class, | xi-vj |2Denotes the distance, V, from the ith sample to the jth cluster centerjIs a clustering center;
the updating of the iterative center and membership is performed by the following relations (5) and (6):
Figure FDA0003349921230000022
Figure FDA0003349921230000023
r represents iteration times, whether | J (r +1) -J (r) | < epsilon is checked, if yes, calculation is stopped, and if not, iteration is continued;
and (3) a subtraction and classification algorithm:
for each point X in the sample set XiThe density value index is calculated according to the following relations (7) and (8):
Figure FDA0003349921230000031
Figure FDA0003349921230000032
selecting the data point x with the highest density indexc1(Density index is D)c1) As a first cluster center; gamma rayaA point-to-point x representing the radius of the area at that point, outside the radiusiThe density index contribution of (a) is small; k-th selected clustering center xck(Density index is D)ck) The density index of each point is corrected by the following relational expression (9):
Figure FDA0003349921230000033
γbis a positive number, typically taken as γb＝1.5γaRepresenting a region in which the density indicator function is significantly reduced;
selecting the point x with the highest density index from the corrected data pointsck+1As a new cluster center; whether the following relation (10) is satisfied or not is judged, and if not, the relation (5) and the relation (6) are returned until the condition is satisfied.
Figure FDA0003349921230000034
5. The method according to claim 1, wherein the selecting the characteristic variables affecting the wind power comprises:
and selecting characteristic variables influencing wind power by adopting a characteristic extraction Relieff algorithm.
6. The method according to claim 5, wherein the selecting the characteristic variables affecting the wind power by using the characteristic extraction Relieff algorithm comprises:
selecting a sample x from the training samplesiSelect k xiThe nearest neighbor sample of (1) is recorded as H;
from xiSelecting k different classes of non-similar nearest neighbor samples to be recorded as M (c);
obtaining a characteristic weight by calculating the intra-class distance and the inter-class distance of the nearest neighbor sample;
by repeating the above process, the weights of all the features are finally obtained.
7. The method according to claim 6, wherein the formula for the Relieff algorithm to update the feature weight is as follows:
Figure FDA0003349921230000041
wherein A is0A feature set that is an original data set; a represents a feature subset of the filtered data set; w [ A ]0]The weight coefficient before updating; w [ A ]]Is the updated weight coefficient; x is the number ofiIs the ith sample, H represents the sum of xiNearest neighbor samples in the same class; diff (A, x)iH) is each feature x in AiAnd H, the calculation formula is as follows:
Figure FDA0003349921230000042
k is the number of nearest neighbors; p (C) is the ratio of the target sample C to the total sample; p (class (x)i) Is x is containediThe ratio of the same type of sample to the total sample; mj (C) represents the jth neighbor sample that is not of the same class as the target sample C; diff (A, x)iMj (C) is each feature x in AiAnd quantitative characterization of the differences between Mj (C).
8. The method according to any one of claims 1 to 7, wherein the building of the wind power prediction model by using the deep neural network based on the characteristic variables comprises:
and constructing the wind power prediction model by adopting a Bi-directional long-short term memory neural network Bi-LSTM based on the characteristic variables.
9. The method according to any one of claims 1 to 7, wherein the evaluating the prediction result of the wind power prediction model comprises:
and evaluating the prediction result of the wind power prediction model through the average absolute error and the root mean square error.
10. A short-term wind power prediction apparatus, the apparatus comprising:
the acquisition module is used for acquiring fan operation data of the wind power plant;
the removing module is used for removing the wind power abnormal value in the operation data;
the selection module is used for selecting characteristic variables influencing wind power;
the establishing module is used for establishing a wind power prediction model by adopting a deep neural network based on the characteristic variables;
and the evaluation module is used for evaluating the prediction result of the wind power prediction model.
CN202111334208.3A 2021-11-11 2021-11-11 Short-term wind power prediction method and device Pending CN114298130A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111334208.3A CN114298130A (en) 2021-11-11 2021-11-11 Short-term wind power prediction method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111334208.3A CN114298130A (en) 2021-11-11 2021-11-11 Short-term wind power prediction method and device

Publications (1)

Publication Number Publication Date
CN114298130A true CN114298130A (en) 2022-04-08

Family

ID=80964666

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111334208.3A Pending CN114298130A (en) 2021-11-11 2021-11-11 Short-term wind power prediction method and device

Country Status (1)

Country Link
CN (1) CN114298130A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108510006A (en) * 2018-04-08 2018-09-07 重庆邮电大学 A kind of analysis of business electrical amount and prediction technique based on data mining
CN110674864A (en) * 2019-09-20 2020-01-10 国网上海市电力公司 Wind power abnormal data identification method with synchronous phasor measurement device
CN112598180A (en) * 2020-12-21 2021-04-02 北京华能新锐控制技术有限公司 Distributed regional wind power prediction method
CN112949945A (en) * 2021-04-15 2021-06-11 河海大学 Wind power ultra-short-term prediction method for improving bidirectional long-short term memory network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108510006A (en) * 2018-04-08 2018-09-07 重庆邮电大学 A kind of analysis of business electrical amount and prediction technique based on data mining
CN110674864A (en) * 2019-09-20 2020-01-10 国网上海市电力公司 Wind power abnormal data identification method with synchronous phasor measurement device
CN112598180A (en) * 2020-12-21 2021-04-02 北京华能新锐控制技术有限公司 Distributed regional wind power prediction method
CN112949945A (en) * 2021-04-15 2021-06-11 河海大学 Wind power ultra-short-term prediction method for improving bidirectional long-short term memory network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
苏舟: ""基于CSO-BP神经网络的电力***短期负荷预测"", 《硕士电子期刊》, 15 February 2018 (2018-02-15), pages 2 *
邢恩恺: ""基于改进VMD预处理与双向LSTM的风功率预测研究"", 《硕士电子期刊》, 15 July 2019 (2019-07-15), pages 2 - 4 *

Similar Documents

Publication Publication Date Title
CN107092582B (en) Online abnormal value detection and confidence evaluation method based on residual posterior
CN111148118B (en) Flow prediction and carrier wave turn-off method and system based on time sequence
CN116757534B (en) Intelligent refrigerator reliability analysis method based on neural training network
CN112557034B (en) Bearing fault diagnosis method based on PCA _ CNNS
CN111723925B (en) Fault diagnosis method, device, equipment and medium for on-road intelligent train air conditioning unit
CN111931983B (en) Precipitation prediction method and system
CN111144552B (en) Multi-index grain quality prediction method and device
CN111768000A (en) Industrial process data modeling method for online adaptive fine-tuning deep learning
CN113012766B (en) Self-adaptive soft measurement modeling method based on online selective integration
CN113723007A (en) Mechanical equipment residual life prediction method based on DRSN and sparrow search optimization BilSTM
CN117349797B (en) Aircraft fault detection method and system based on artificial intelligence
CN112098878A (en) End-to-end neural network establishment method for lithium battery SOH estimation and RUL prediction
CN114015825A (en) Method for monitoring abnormal state of blast furnace heat load based on attention mechanism
CN114298134A (en) Wind power prediction method and device and electronic equipment
CN111507365A (en) Confidence rule automatic generation method based on fuzzy clustering
CN113988415B (en) Medium-and-long-term power load prediction method
CN115062764B (en) Intelligent illuminance adjustment and environmental parameter Internet of things big data system
CN114298130A (en) Short-term wind power prediction method and device
CN115794805A (en) Medium-low voltage distribution network measurement data supplementing method
CN112966449B (en) Manufacturing process quality defect root cause diagnosis method based on variable selection and probabilistic neural network
CN113126489B (en) Intelligent BIT design method of heavy-duty gas turbine control system based on CNN-GRU-BINN
CN109902762A (en) The data preprocessing method deviateed based on 1/2 similarity
CN112685933B (en) Method for predicting residual service life of roller screw pair
CN118133086A (en) Food quality classification method based on microbial growth prediction
CN114417248A (en) DCGAN-based method and system for monitoring abnormal quality of linear contour process

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