CN114865620B - Wind power plant generating capacity prediction method based on machine learning algorithm - Google Patents
Wind power plant generating capacity prediction method based on machine learning algorithm Download PDFInfo
- Publication number
- CN114865620B CN114865620B CN202210467269.5A CN202210467269A CN114865620B CN 114865620 B CN114865620 B CN 114865620B CN 202210467269 A CN202210467269 A CN 202210467269A CN 114865620 B CN114865620 B CN 114865620B
- Authority
- CN
- China
- Prior art keywords
- vector
- dimension
- attention
- wind power
- network
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000010801 machine learning Methods 0.000 title claims abstract description 14
- 239000013598 vector Substances 0.000 claims abstract description 96
- 238000010248 power generation Methods 0.000 claims abstract description 47
- 230000007246 mechanism Effects 0.000 claims abstract description 12
- 238000012216 screening Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 22
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000006243 chemical reaction Methods 0.000 claims description 11
- 239000000126 substance Substances 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 241001522296 Erithacus rubecula Species 0.000 claims description 3
- 230000009471 action Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000012512 characterization method Methods 0.000 claims description 3
- 238000005315 distribution function Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000004044 response Effects 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 claims 5
- 238000011017 operating method Methods 0.000 claims 3
- 108090000672 Annexin A5 Proteins 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Entrepreneurship & Innovation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Power Engineering (AREA)
- Biophysics (AREA)
- Development Economics (AREA)
- Molecular Biology (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wind power plant generating capacity prediction method based on a machine learning algorithm, which comprises the following steps: obtaining and inputting meteorological historical data to obtain a vector capable of representing meteorological data characteristics, training and learning characteristics related to a time sequence among variables in a meteorological data characteristic sequence by using a circulating high-speed network, re-screening the characteristic vector in different dimensions by using a circulating high-speed path network encoder and a multi-layer space-time attention mechanism to respectively obtain a time dimension attention vector and a network layer dimension attention vector, obtaining and inputting wind power generation capacity historical data, obtaining a prediction result of wind power generation capacity by performing full link layer dimension matching through decoding operation of a circulating high-speed path network decoder, and calculating a confidence interval of the wind power generation capacity prediction result. The method not only effectively improves the wind power generation prediction precision, but also can provide confidence interval information of wind power generation prediction, and enriches the decision space of a power grid manager.
Description
Technical Field
The invention belongs to the technical field of wind power generation capacity prediction, and particularly relates to a wind power generation field power generation capacity prediction method based on a machine learning algorithm.
Background
The generated energy of the wind power plant has strong instability under the influence of meteorological conditions such as wind speed and air pressure. The prediction of the wind power generation amount in the unit of hours or days is significant for the comprehensive scheduling of a power grid and the operation or maintenance of a wind turbine generator, but has extremely high difficulty. The machine learning algorithm can effectively represent deep features of input data, and is widely applied to prediction of wind power generation. However, performance degradation is caused by the phenomena of input sequence information loss, gradient disappearance caused by network layer superposition and the like in machine learning, particularly in deep learning algorithm application, and therefore the power training efficiency and the prediction accuracy are influenced. Meanwhile, the existing prediction method does not provide confidence interval distribution information of a prediction result on the basis of more accurately predicting the wind power generation amount, and the method is more important for more efficient power grid decision.
Disclosure of Invention
Aiming at the problems in the prior art, the invention aims to provide a method for predicting the power generation amount of a wind power plant based on a machine learning algorithm.
In order to achieve the purpose, the method is based on a circulating high-speed path network and combines a multi-layer space-time attention mechanism to predict the wind power generation amount and further calculate the probability distribution of the wind power generation amount. The adopted technical scheme comprises the following steps:
step 1, acquiring and inputting meteorological historical data to obtain a vector capable of representing meteorological data characteristics:
step 1.1, acquiring meteorological historical data to obtain a meteorological historical data sequence which can be input into a convolutional neural network for feature extraction: (x) 1 ,x 1 ,…,x t ,…,x T-1 )
Wherein T is equal to {1,2,. Multidot., T-1}, x t ∈R n N-dimensional meteorological real vector data at the time T, wherein T is a time sequence of a target to be predicted;
step 1.2, inputting the sequences into a convolutional neural network, and performing convolutional operation to obtain a sequence capable of representing meteorological data characteristics:
(w 1 ,w 2 ,…,w t ,…,w T-1 )
wherein, w t ∈R m Representing meteorological characteristic real vector data for dimension m at t moment after processing;
step 2, training and learning the characteristics related to the time sequence among variables in the meteorological data characteristic sequence by using a circulating high-speed network: with S '= S' (W, W) S' )、t'=T'(w,W T' ) And C '= C' (W, W) C' ) Respectively representing the outputs, w, after action conversion by nonlinear conversion tanh function S ', sigmoid function T' and sigmoid function C [t] For the meteorological feature vector input at the time t, the hidden state in the circulating high-speed path network is updated to
Wherein the content of the first and second substances,the hidden state vector of the dimension I is output by the circulating expressway network unit at the time point t of the K layer, K belongs to {1, 2.. K }, and K is the number of network layers and is specified when K =0
Wherein, W S' ,W T' ,W C' ∈R l×m 、R S'k ,R C'k ,R T'k ∈R l×l And b Hk ,b C'k ,b T'k ∈R l A weight matrix and a bias unit respectively representing the S ', T ' and C ' conversion of the k-th layer, and an indicator function I {. Cndot. } representing a feature vector w [t] Only at layer 1 (k = 1) of the round robin highway network participates in the operation,indicating that all of the original input information is retained,representing conversion of all input information;
and 3, re-screening the feature vectors in different dimensions through the encoding operation of an encoder and a multi-layer space-time attention mechanism to respectively obtain a time dimension attention vector and a network hierarchy dimension attention vector:
step 3.1, screening different coding characteristics by adopting a space-time attention mechanism to obtain a time dimension attention vector:
let the decoder have a hidden state vector of the k-th layer at the time T-1To D T-1 The Query vector Query1 is obtained after the deformation operation, and the Query vector Query and the hidden vector of the encoder at the kth layer at the T (T is more than 1 and less than or equal to T-1) timeThe attention weight between is expressed as
Wherein, query1 belongs to R p ,V k ∈R l ,T' k ∈R l×p ,U k ∈R l×l ,V k 、T’ k And U k Respectively are k-th layer nonlinear transformation matrixes, p is a query vector dimension, and the attention weight after normalization processing is expressed as
The k-th layer time-dimension local attention vector can be obtained
The local attention vectors of each layer are spliced to obtain the attention vector of time dimension
And 3.2, screening different coding characteristics by adopting a space-time attention mechanism to obtain a network level dimension attention vector:
for the decoder at T-1Hidden state vector of etch/k layerAfter reshape operation, the Query vector Query2 is obtained, which is obtained from the operation of reshapeThe attention weight between is expressed as
Wherein Query2 ∈ R p ,V t ∈R l ,T' t ∈R l×p ,U t ∈R l×l ,V t 、T’ t And U t Respectively, the t-th time nonlinear transformation matrix is normalized and the attention weight is expressed as
The network hierarchy dimension local attention vector at the t moment can be obtained
The local attention vectors of all layers are spliced to obtain the network level dimension attention vector
And 4, acquiring and inputting historical data of the wind power generation capacity, and obtaining a prediction result of the wind power generation capacity through decoding operation of a decoder and dimension matching of a full link layer:
step 4.1, obtaining and inputting historical data of wind power generation amount to obtain a power generation amount data sequence:
(y 1 ,y 2 ,…,y t ,…,y T-1 )
step 4.2, performing dimension matching on the time dimension and the network dimension attention vector and the wind power generation capacity historical data through the full connection layer to obtain a characterization vector
Wherein the content of the first and second substances,andrepresenting a full connection layer weight matrix, d being the weight matrix dimension,is a bias unit;
and 4.3, updating the hidden state vector of the encoder:
subjecting the obtained mixture toAs an input to the k-th loop freeway network, an encoder state vector is updated to
Wherein the content of the first and second substances,
wherein, the first and the second end of the pipe are connected with each other,respectively representing the weight matrices of different layers of the encoder stage cyclic highway network,representing different bias units, q is the weight matrix dimension,
and 4.4, obtaining and outputting a prediction result of the wind power generation amount at the T moment:
wherein the content of the first and second substances,a hidden state vector of a K-th layer of a decoder at the moment T-1 is represented by W, V and H, and b is a bias unit;
step 5, calculating a confidence interval of the wind power generation amount prediction result:
if the distribution function of the random variable Y satisfies F (Y) = P (Y ≦ Y), the τ -th quantile thereof may be defined as
Q(τ)=inf{y:F(y)≥τ},τ∈(0,1)
The fractal regression model is optimized by using pinball loss function minimum as an index through a neural network back propagation method
Wherein N is the predicted horizontal number of quantiles, X i (i =1,2, \8230;, N) is a sample of the density function f (x),the output values of different quantiles. The influence of the interpretation variables on the condition quantiles of the response variables at different quantiles can be measured by continuously adjusting the values of W and b in the learning process. Obtaining the optimal parameter vectorAndthen, the optimum estimated value of Y is
The quantiles of the wind power generation capacity predicted value under different quantiles are used as input of a Gaussian kernel, and the probability distribution estimation value is obtained by selecting a proper window width
Wherein h is the window width, and the Gaussian kernel function K (-) is expressed as
The invention has the beneficial effects that: the machine learning, particularly the deep learning algorithm, has advantages in short-term prediction in unit of hour or day on the wind power generation amount with strong instability; in order to further improve the short-term prediction precision and the prediction performance of the wind power generation, a circulating high-speed path network and a multi-layer space-time attention mechanism are introduced, so that the characteristics of a learning data sequence are better represented, the loss of input sequence information is reduced, and the information selection and utilization capacity is improved; in order to provide more and more valuable decision information about the prediction result, a complete probability distribution interval of a short-term wind power generation capacity prediction value is obtained by combining quantile regression and a Gaussian kernel function method.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
Detailed Description
The invention will be further described with reference to the drawings attached to the description, but the scope of the invention is not limited thereto.
Step 1, acquiring and inputting meteorological historical data to obtain a vector capable of representing meteorological data characteristics:
step 1.1, acquiring meteorological historical data to obtain a meteorological historical data sequence which can be input into a convolutional neural network for feature extraction: (x) 1 ,x 1 ,…,x t ,…,x T-1 )
Wherein, T is an element of {1,2,. Eta., T-1}, x t ∈R n N-dimensional meteorological real vector data at the moment T, wherein T is a time sequence of a target to be predicted;
the meteorological historical data mainly comprises 2-meter wind speed, 10-meter wind speed, 50-meter wind speed, roughness, ground level radiation, atmospheric level radiation, air temperature, air density, air pressure, on-shore wind speed profile, off-shore wind speed profile and the like, and the table 1 shows the meteorological historical data of each meteorological point per hour:
TABLE 1 weather History data sequence
Step 1.2, inputting the sequences into a convolutional neural network, and obtaining the sequences capable of representing meteorological data characteristics after convolution operation processing: (w) 1 ,w 2 ,…,w t ,…,w T-1 )
Wherein w t ∈R m Representing real vector data of meteorological features for m-dimension at t moment after processing;
after convolution operation, a sequence representing the meteorological historical data characteristics is obtained, and a table 2 shows the meteorological characteristic data sequence:
TABLE 2 weather characteristic data series
Step 2, training and learning the characteristics related to the time sequence among variables in the meteorological data characteristic sequence by using a circulating high-speed network: with S '= S' (W, W) S' )、t'=T'(w,W T' ) And C '= C' (W, W) C' ) Respectively representing the outputs, w, after action conversion by nonlinear conversion tanh function S ', sigmoid function T' and sigmoid function C [t] For the meteorological characteristic vector input at the time t, the hidden state in the circulating high-speed path network is updated to
Wherein the content of the first and second substances,the hidden state vector of the dimension I is output by the circulating expressway network unit at the time point t of the K layer, K belongs to {1, 2.. K }, and K is the number of network layers and is specified when K =0
Wherein, W S' ,W T' ,W C' ∈R l×m 、R S'k ,R C'k ,R T'k ∈R l×l And b Hk ,b C'k ,b T'k ∈R l A weight matrix and a bias unit respectively representing the conversion of the k-th layer S ', T ' and C ', and an indication function I {. DEG } representing a characteristic vector w [t] Only at layer 1 (k = 1) of the round robin highway network participates in the operation,indicating that all of the original input information is retained,representing conversion of all input information;
taking the data from 1 month and 1 day to 10 months and 17 days as the training set of the model, taking the data from 10 months and 18 days to 11 months and 23 days as the test set of the model, setting the relevant parameters, and obtaining the updated internal hidden state, as shown in table 3:
TABLE 3 updated hidden State vector List
And 3, re-screening the feature vectors in different dimensions through the encoding operation of an encoder and a multi-layer space-time attention mechanism to respectively obtain a time dimension attention vector and a network hierarchy dimension attention vector:
step 3.1, screening different coding characteristics by adopting a space-time attention mechanism to obtain a time dimension attention vector:
let the decoder hide the state of the k-th layer at time T-1Vector isTo D T-1 After reshape operation, the Query vector Query1 is obtained, which is associated with the hidden vector of the k-th layer at the T (T is more than 1 and less than or equal to T-1) time of the encoderThe attention weight between is expressed as
Wherein, query1 belongs to R p ,V k ∈R l ,T' k ∈R l×p ,U k ∈R l×l ,V k 、T’ k And U k Respectively, k-th layer nonlinear transformation matrix, p is query vector dimension, and after normalization treatment, attention weight is expressed as
The k-th layer time-dimension local attention vector can be obtained
The local attention vectors of each layer are spliced to obtain the attention vector of time dimension
Table 4 is the time-dimensional attention vector after the stitching process:
TABLE 4 time-dimensional attention vector
Step 3.2, screening different coding characteristics by adopting a space-time attention mechanism to obtain a network level dimension attention vector:
hidden state vector of k layer at T-1 time of decoderGet Query vector Query2 after reshape operation, it andthe attention weight between is expressed as
Wherein Query2 ∈ R p ,V t ∈R l ,T' t ∈R l×p ,U t ∈R l×l ,V t 、T’ t And U t Respectively, the t-th time nonlinear transformation matrix, and the attention weight after normalization treatment is expressed as
The network hierarchy dimension local attention vector at the t moment can be obtained
The local attention vectors of all layers are spliced to obtain the network level dimension attention vector
Table 5 shows the attention vectors after the local attention items of each layer of network hierarchy are spliced:
TABLE 5 network hierarchy dimension attention vector
Index | 0 | 1 | 2 | 3 | 4 | …… | 253 | 254 | 255 |
0 | 0.1864 | 0.1060 | -0.1316 | -0.0669 | 0.0009 | …… | 0.1204 | 0.0836 | -0.0913 |
1 | 0.1980 | 0.0854 | -0.1102 | -0.0839 | -0.0078 | …… | 0.1217 | 0.0739 | -0.1037 |
…… | …… | …… | …… | …… | …… | …… | …… | …… | …… |
334 | 0.1997 | 0.0821 | -0.1207 | -0.0778 | -0.0080 | …… | 0.1180 | 0.0839 | -0.1028 |
335 | 0.2072 | 0.0967 | -0.0942 | -0.0817 | -0.0165 | …… | 0.1171 | 0.0812 | -0.1058 |
And 4, acquiring and inputting historical data of the wind power generation capacity, and obtaining a prediction result of the wind power generation capacity through decoding operation of a decoder and dimension matching of a full link layer:
step 4.1, obtaining and inputting historical data of wind power generation amount to obtain a power generation amount data sequence:
(y 1 ,y 2 ,…,y t ,…,y T-1 )
the historical data of the wind power generation amount is 8784 sample data in each hour at intervals and in a time span of one year, and the table 6 shows part of the sample data:
TABLE 6 historical data List of wind power generation
Step 4.2, performing dimension matching on the time dimension and the network dimension attention vector and the wind power generation capacity historical data through the full connection layer to obtain a characterization vector
Wherein, the first and the second end of the pipe are connected with each other,andrepresenting the fully connected layer weight matrix, d is the weight matrix dimension,is a bias unit;
and 4.3, updating the hidden state vector of the encoder:
subjecting the obtained product toAs an input to the k-th loop freeway network, an encoder state vector is updated to
Wherein the content of the first and second substances,
wherein, the first and the second end of the pipe are connected with each other,respectively representing the weight matrices of different layers of the encoder stage circular high-speed path network,representing different bias units, and q is a weight matrix dimension;
and 4.4, obtaining and outputting a prediction result of the wind power generation amount at the T moment:
wherein the content of the first and second substances,a hidden state vector of a K-th layer of a decoder at the moment T-1 is represented by W, V and H, and b is a bias unit;
the prediction results of the prediction method at the time T and different network layers can be obtained by respectively adopting the average absolute percentage error MAPE and the root mean square error RMSE methods, the point prediction results are shown in the table 7, and the point prediction results are divided into the point prediction results shown in the table 8:
TABLE 7 prediction results
TABLE 8 Point prediction results
Step 5, calculating a confidence interval of the wind power generation amount prediction result:
if the distribution function of the random variable Y satisfies F (Y) = P (Y ≦ Y), the τ th quantile can be defined as
Q(τ)=inf{y:F(y)≥τ},τ∈(0,1)
The fractal regression model is optimized by using pinball loss function minimum as an index through a neural network back propagation method
Wherein N is the predicted horizontal number of quantiles, X i (i =1,2, \8230;, N) is a sample of the density function f (x),the output values of different quantiles. The influence of the interpretation variables on the condition quantiles of the response variables at different quantiles can be measured by continuously adjusting the values of W and b in the learning process. Obtaining the optimal parameter vectorAndthen, the optimum estimate of Y is
The quantiles of the wind power generation capacity predicted value under different quantiles are used as the input of a Gaussian kernel, and the probability distribution estimation value can be obtained by selecting a proper window width
Wherein h is the window width, and the Gaussian kernel function K (-) is expressed as
Selecting interval coverage (PICP) and average width of Prediction Interval (PINAW) as evaluation indexes of interval prediction, predicting the wind power generation amount of a complete time interval from 11 months 3 days to 11 months 10 days in 1 week, respectively selecting 85%, 90% and 95% confidence levels to establish the prediction interval of the wind power generation amount, and the result is shown in Table 9:
TABLE 9 comparison of interval prediction indexes
Confidence level | PICP | PINAW |
85% | 0.997 | 0.078 |
90% | 1 | 0.097 |
95% | 1 | 0.116 |
From table 9, it can be seen that the algorithm obtains a relatively high PICP and a relatively small PINAW at 85%, 90%, and 95% confidence levels, which indicates that the algorithm can provide interval distribution information of wind power generation pre-measurement in a relatively small range and a relatively large degree, and provides an optimization space for power grid decision.
Claims (6)
1. The wind power plant power generation amount prediction method based on the machine learning algorithm is characterized by comprising the following steps of:
step 1: acquiring and inputting meteorological historical data to obtain a vector capable of representing meteorological data characteristics;
and 2, step: training and learning characteristics related to a time sequence among variables in a meteorological data characteristic sequence by using a circulating high-speed network;
and step 3: re-screening the feature vectors in different dimensions through the encoding operation of a circulating high-speed path network encoder and a multi-layer space-time attention mechanism to respectively obtain a time dimension attention vector and a network hierarchy dimension attention vector;
and 4, step 4: acquiring and inputting historical data of wind power generation capacity, and acquiring a prediction result of the wind power generation capacity through decoding operation of a circulating high-speed path network decoder and dimension matching of a full link layer;
and 5: and calculating a confidence interval of the wind power generation amount prediction result.
2. A method for predicting the power production of a wind farm based on a machine learning algorithm according to claim 1, characterized in that the specific operating procedure of step 1 comprises the following steps:
step 1.1, acquiring meteorological historical data to obtain a meteorological historical data sequence which can be input into a convolutional neural network for feature extraction: (x) 1 ,x 1 ,…,x t ,…,x T-1 ) Wherein, T is an element {1,2, 1}, x t ∈R n N-dimensional meteorological real vector data at the time T, wherein T is a time sequence of a target to be predicted;
step 1.2, inputting the sequences into a convolutional neural network, and obtaining the sequences (w) capable of representing meteorological data characteristics after convolutional operation processing 1 ,w 2 ,…,w t ,…,w T-1 ) Wherein w is t ∈R m And representing real vector data of meteorological features for m-dimension at the t moment after processing.
3. The machine learning algorithm based wind farm energy production prediction method according to claim 2, characterized in that the specific operation of step 2 comprises the steps of: with S '= S' (W, W) S' )、t'=T'(w,W T' ) And C '= C' (W, W) C' ) Respectively representing the outputs after action conversion by nonlinear conversion tanh function S ', sigmoid function T ' and sigmoid function C ', w [t] For the meteorological characteristic vector input at the time t, the hidden state in the circulating high-speed path network is updated to
Wherein the content of the first and second substances,and the hidden state vector is expressed by a dimension I output by the circulating expressway network unit at the time point t of the K layer, K belongs to {1, 2.. K }, and K is the number of network layers and is specified when K =0
Wherein, W S' ,W T' ,W C' ∈R l×m 、R S'k ,R C'k ,R T'k ∈R l×l And b Hk ,b C'k ,b T'k ∈R l A weight matrix and a bias unit respectively representing the conversion of the k-th layer S ', T ' and C ', and an indication function I {. DEG } representing a characteristic vector w [t] Only at layer 1 of the round robin highway network,indicating that all of the original input information is retained,indicating that all input information is converted.
4. A method for predicting the power production of a wind farm based on a machine learning algorithm according to claim 3, characterized in that the specific operating procedure of step 3 comprises the following steps:
step 3.1, screening different coding characteristics by adopting a space-time attention mechanism to obtain a time dimension attention vector:
let the decoder at the T-1 time point and the hidden state vector of the k layer beTo D T-1 After the deformation operation, the Query vector Query1 is obtained, which is the hidden vector of the k layer at the t time point with the encoderThe attention weight between is expressed as
Wherein, query1 belongs to R p ,V k ∈R l ,T' k ∈R l×p ,U k ∈R l×l ,V k 、T’ k And U k Respectively are k-th layer nonlinear transformation matrixes, p is a query vector dimension, and the attention weight after normalization processing is expressed as
Get the k-th layer time-dimensional local attention vector as
The local attention vectors of each layer are spliced to obtain the attention vector with time dimension as
Step 3.2, screening different coding characteristics by adopting a space-time attention mechanism to obtain a network level dimension attention vector:
hidden state vector of k layer at time T-1 of decoderAfter reshape operation, the Query vector Query2 is obtained, which is obtained from the operation of reshapeIs expressed as the attention weight between
Wherein Query2 ∈ R p ,V t ∈R l ,T' t ∈R l×p ,U t ∈R l×l ,V t 、T’ t And U t Respectively, the t-th time nonlinear transformation matrix is normalized and the attention weight is expressed as
The network hierarchy dimension local attention vector at the t-th moment can be obtained as
The local attention vectors of each layer are spliced to obtain the network level dimension attention vector of
5. A method for predicting the power production of a wind farm based on machine learning algorithms according to claim 4, characterized in that the specific operating procedure of step 4 comprises the following steps:
step 4.1, obtaining and inputting historical data of wind power generation capacity to obtain a power generation capacity data sequence:
(y 1 ,y 2 ,…,y t ,…,y T-1 )
step 4.2, performing dimension matching on the time dimension and the network dimension attention vector and the wind power generation capacity historical data through the full connection layer to obtain a characterization vector
Wherein the content of the first and second substances,andrepresenting a full connection layer weight matrix, d being the weight matrix dimension,is a bias unit;
and 4.3, updating the hidden state vector of the encoder:
subjecting the obtained product toAs an input to the k-th loop freeway network, the encoder state vector is updated to
Wherein, the first and the second end of the pipe are connected with each other,
wherein the content of the first and second substances,respectively representing the weight matrices of different layers of the encoder stage cyclic highway network,representing different bias units, q is the weight matrix dimension,
and 4.4, obtaining and outputting a prediction result of the wind power generation amount at the T moment:
6. The machine learning algorithm based wind farm energy production prediction method according to claim 5, characterized in that the specific operation process of step 5 is that if the distribution function of the random variable Y satisfies F (Y) = P (Y ≦ Y), the τ th quantile thereof is defined as
Q(τ)=inf{y:F(y)≥τ},τ∈(0,1)
Optimizing the quantile regression model by using the minimum pinball loss function as an index through a neural network back propagation method
Wherein N is the number of predicted quantile levels, X i Is a sample of the density function f (x), i =1,2, \8230;, N,for the output values of different quantiles, the influence of the explanatory variables on the quantiles of the response variable condition at different quantiles is measured and calculated by continuously adjusting the values of W and b in the learning process to obtain the optimal parameter vectorAndthen, the optimum estimated value of Y is
The quantiles of the wind power generation capacity predicted value under different quantiles are used as the input of a Gaussian kernel, and the probability distribution estimated as
Wherein h is the window width, and the Gaussian kernel function K (-) is expressed as
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210467269.5A CN114865620B (en) | 2022-04-29 | 2022-04-29 | Wind power plant generating capacity prediction method based on machine learning algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210467269.5A CN114865620B (en) | 2022-04-29 | 2022-04-29 | Wind power plant generating capacity prediction method based on machine learning algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114865620A CN114865620A (en) | 2022-08-05 |
CN114865620B true CN114865620B (en) | 2023-01-10 |
Family
ID=82635746
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210467269.5A Active CN114865620B (en) | 2022-04-29 | 2022-04-29 | Wind power plant generating capacity prediction method based on machine learning algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114865620B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115907233B (en) * | 2023-01-06 | 2023-05-09 | 北京科技大学 | Wind power generation quantile prediction method based on machine mental model and self-attention |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107766937A (en) * | 2017-09-11 | 2018-03-06 | 重庆大学 | Feature based chooses and the wind power ultra-short term prediction method of Recognition with Recurrent Neural Network |
CN111626506B (en) * | 2020-05-27 | 2022-08-26 | 华北电力大学 | Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof |
CN113988421A (en) * | 2021-10-29 | 2022-01-28 | 北京中恒博瑞数字电力科技有限公司 | Wind turbine generator power intelligent prediction method based on deep learning |
CN114298132A (en) * | 2021-11-11 | 2022-04-08 | 华能新能源股份有限公司 | Wind power prediction method and device and electronic equipment |
CN114282711A (en) * | 2021-12-03 | 2022-04-05 | 中国电建集团贵州电力设计研究院有限公司 | Photovoltaic short-term power generation capacity prediction method integrated with time-frequency analysis |
-
2022
- 2022-04-29 CN CN202210467269.5A patent/CN114865620B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN114865620A (en) | 2022-08-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110414788B (en) | Electric energy quality prediction method based on similar days and improved LSTM | |
CN111697621B (en) | Short-term wind power prediction method based on EWT-PDBN combination | |
CN112288164B (en) | Wind power combined prediction method considering spatial correlation and correcting numerical weather forecast | |
CN113705877B (en) | Real-time moon runoff forecasting method based on deep learning model | |
CN109599872B (en) | Power system probability load flow calculation method based on stack noise reduction automatic encoder | |
CN112116162B (en) | Power transmission line icing thickness prediction method based on CEEMDAN-QFAOA-LSTM | |
CN110717610B (en) | Wind power prediction method based on data mining | |
CN113344288B (en) | Cascade hydropower station group water level prediction method and device and computer readable storage medium | |
CN114865620B (en) | Wind power plant generating capacity prediction method based on machine learning algorithm | |
CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
CN114676923A (en) | Method and device for predicting generated power, computer equipment and storage medium | |
CN114897129A (en) | Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM | |
CN116702937A (en) | Photovoltaic output day-ahead prediction method based on K-means mean value clustering and BP neural network optimization | |
CN116522124A (en) | Dissolved oxygen content prediction method and system based on influence of environmental factors | |
CN112288157A (en) | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning | |
CN115907195A (en) | Photovoltaic power generation power prediction method, system, electronic device and medium | |
CN115169742A (en) | Short-term wind power generation power prediction method | |
CN114429238A (en) | Wind turbine generator fault early warning method based on space-time feature extraction | |
CN112836876B (en) | Power distribution network line load prediction method based on deep learning | |
CN112507613B (en) | Second-level ultra-short-term photovoltaic power prediction method | |
CN112669168B (en) | Short-term prediction method for wind power | |
CN110276478B (en) | Short-term wind power prediction method based on segmented ant colony algorithm optimization SVM | |
CN113095547A (en) | Short-term wind power prediction method based on GRA-LSTM-ICE model | |
CN111506868B (en) | Ultra-short-term wind speed prediction method based on HHT weight optimization | |
CN117096860A (en) | Overhead transmission line current-carrying capacity interval prediction method and equipment based on LSSVM model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |