CN111985711B - Wind power probability prediction model building method - Google Patents
Wind power probability prediction model building method Download PDFInfo
- Publication number
- CN111985711B CN111985711B CN202010834929.XA CN202010834929A CN111985711B CN 111985711 B CN111985711 B CN 111985711B CN 202010834929 A CN202010834929 A CN 202010834929A CN 111985711 B CN111985711 B CN 111985711B
- Authority
- CN
- China
- Prior art keywords
- wind power
- model
- meta
- data set
- parameter vector
- 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 31
- 239000013598 vector Substances 0.000 claims abstract description 50
- 238000012549 training Methods 0.000 claims abstract description 47
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 238000002156 mixing Methods 0.000 claims abstract description 12
- 230000002159 abnormal effect Effects 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000010248 power generation Methods 0.000 description 3
- 238000009472 formulation Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Primary Health Care (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Power Engineering (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wind power probability prediction model building method, which comprises the following steps: removing abnormal values in the initial data set, and selecting weather variables with the relevance to the wind power greater than a preset threshold value as a training data set of a wind power probability prediction model based on a gray relevance theory; establishing an improved natural gradient lifting meta-model: updating the parameter vector; and carrying out Blending model fusion on the improved natural gradient lifting meta-models, establishing a new meta-model for training, and outputting a final prediction statistical parameter vector. The wind power probability prediction model establishment method provided by the invention can provide complete wind power uncertainty information, has higher prediction area coverage rate and smaller prediction interval average width ratio, and provides more accurate reference for constructing a high-efficiency intelligent new energy power system.
Description
Technical Field
The invention relates to the technical field of wind power prediction of wind power plants, in particular to a wind power probability prediction model building method.
Background
With the low-carbon development of energy structures, the permeability of renewable energy power grids represented by wind power is increased year by year. However, due to the strong randomness of wind energy, the wind power has serious fluctuation, the complete uncertainty information can not be obtained only by carrying out point prediction on the wind power, and the safe and stable operation of the power grid is challenging. Therefore, in order to better utilize wind power generation, a scheduling plan is effectively adjusted, the wind power bidding grid-connected advantage is enlarged, and accurate probability prediction of wind power is critical. However, how the probability prediction model should be built, how the model input variables should be selected, and how the model should be optimized to improve the prediction ability and generalization ability have not been well defined.
Disclosure of Invention
The invention aims to provide a wind power probability prediction model building method, which can predict and acquire uncertainty information of wind power, is beneficial to improving the operation reliability of a wind power grid and solves the problems of low wind power generation consumption level and difficult scheduling plan formulation caused by the lack of uncertainty information of wind power.
In order to achieve the above object, the present invention provides the following solutions:
a wind power probability prediction model building method comprises the following steps:
s1, data preprocessing: removing abnormal values in the initial data set, and selecting weather variables with the relevance to the wind power greater than a preset threshold value as a training data set of a wind power probability prediction model based on a gray relevance theory;
s2, establishing an improved natural gradient lifting meta-model: based on a training data set, predicting parameter vectors of wind power probability distribution, establishing a connection between a general gradient and a natural gradient through Fisher information quantity, selecting classification and regression trees as a base learner, and establishing an improved natural gradient lifting meta-model to realize parameter vector updating;
s3, blending model fusion: and carrying out Blending model fusion on the improved natural gradient lifting meta-models, establishing a new meta-model for training, and outputting a final prediction statistical parameter vector.
Optionally, in the step S1, abnormal values in the initial data set are removed, and based on a gray correlation theory, a meteorological variable with a degree of correlation with wind power greater than a preset threshold is selected as a training data set of a wind power probability prediction model, which specifically includes:
s101, eliminating abnormal values in initial data by using a box graph;
s102, using wind power as a reference data column and related meteorological variables as a comparison data column, performing initialization processing on each sequence, calculating association coefficients based on a gray association theory to represent association degrees of two groups of sequences, and selecting meteorological variables with association degrees larger than a threshold value as a training data set of a prediction model.
Optionally, in step S101, the removing the outlier in the initial data by using the bin graph specifically includes:
the outlier cutoff upper and lower limits are determined by equation (1):
wherein: min and max represent the upper limit and the lower limit of data interception; q (Q) 1 、Q 3 Respectively representing upper and lower quartiles; iqr=q 3 -Q 1 。
Optionally, in step S102, wind power is taken as a reference data column, relevant weather variables are taken as a comparison data column, each sequence is initialized, a correlation coefficient is calculated to represent the correlation degree of two groups of sequences based on a gray correlation theory, and weather variables with the correlation degree larger than a threshold value are selected as training data sets of a prediction model, which specifically includes:
1) Normalizing the time series of each variable, and taking the kth of the n meteorological variable series as a comparison series S k (t) the wind power sequence is a reference sequence S 0 (t) calculating the difference between the two as the absolute value sequence delta k (t) as shown in formula (2), wherein k E (1, n),
Δ k (t)＝|S k (t)-S 0 (t)| (2)
2) Calculating the correlation coefficient eta k (t)：
Wherein: min (·) and Max (·) represent minimum and maximum values of the sequence, and ρ is a resolution coefficient;
3) Solving the association degree gamma k ：
Wherein: t (T) n Is the sequence length;
4) Setting a threshold valueAnd selecting meteorological variables with the association degree larger than a threshold value to form a training data set.
Optionally, in the step S2, based on the training data set, a parameter vector of wind power probability distribution is predicted, a relationship is established between a general gradient and a natural gradient through Fisher information, a classification tree and a regression tree are selected as a base learner, and an improved natural gradient lifting meta-model is established, so as to realize parameter vector update, and the method specifically includes:
s201, let D include n D Samples, m features, i.e. d= { (x) i ，y i )}(x i ∈R m ，y i E R), where x i Characterizing the eigenvector of the ith sample, y i Representing the label value corresponding to the ith sample, i E (1, n) D )；
S202, y i Establishes a scoring function S (theta, y) based on the shannon information amount i )：
S(θ,y i )＝-log P θ (y i ) (8)
Wherein: p (P) θ (y i ) Is y i Probability values in the predictive probability distribution; θ is a parameter vector of the predictive probability distribution;
s203, let-log P θ (y i ) F (θ), taylor expansion is performed on f (θ+d'), and the third order and above remainder are discarded:
wherein: d' is theta edgeMoving an infinitely small step vector; />Representing a natural gradient;
converting the European space into a statistical manifold, and processing the formula (12) under the Riemann space:
wherein the calculation of the primary term can be simplified to:
the remainder is denoted as:
wherein: psi (θ), a Riemann metric of statistical manifold at θ, is used to characterize P θ (yi) the Fisher information amount brought by:
thereby realizing the calculation of natural gradient through general gradient
S204, selecting classification and regression trees as base learners, and generating a group of new base learners along the generalized natural gradient direction by each iteration of the base learners, thereby establishing an improved natural gradient lifting meta-model;
s205, updating a parameter vector theta of wind power probability distribution:
wherein: θ 0 As an initial parameter vector, α m Is a scale factor, beta is a unified learning rate, B m Is a unified representation of the base learner.
Optionally, in the step S3, blending the multiple improved natural gradient lifting meta-models, building a new meta-model for training, so as to output a final predicted statistical parameter vector, which specifically includes:
s301, initial dataset segmentation: dividing an original training data set into a sub training set DT and a test set DA according to a proportion, and defining an original prediction data set as DP;
s302, model fusion:
given the confidence level, building V NGBoost meta-models MO 1 、MO 2 、…、MO V Learning DT by using the metamodels, and outputting a predicted result DA_ P, DP _P of DA and DP on the metamodels after training is completed, wherein DA_ P, DP _P is an initial statistical parameter vector of the predicted values corresponding to DA and DP;
the predicted mean value determined by DA_P and the actual result DA_OUT corresponding to the original DA data form a new data set, and a new meta-model MO is established DA Training and obtaining predicted output MO DA P, where MO is DA P is the corrected predicted statistical parameter vector;
MO is prepared from DA P and DP P form new data set, and new meta-model MO is built P Training is carried out, so that a final prediction statistical parameter vector is output, the upper limit and the lower limit of a predicted value under a given confidence level are calculated through the vector, and the points are connected into a predicted value upper limit curve and a predicted value lower limit curve.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: according to the wind power probability prediction model establishment method provided by the invention, a natural gradient lifting meta-model of generalized natural gradient calculation is utilized; compared with the traditional point prediction model, the model provided by the invention solves the application defect of the traditional Boosting algorithm in solving the problem of wind power probability prediction, has higher generalization and robustness, has higher prediction area coverage rate and smaller average width occupation ratio of a prediction interval, has the reinforced prediction effect, and simultaneously has lower model redundancy, and the model building method is more suitable for practical engineering application, thereby being beneficial to improving the power grid scheduling economy and the running safety of a wind power plant with renewable energy grid connection.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is an initial data diagram of an embodiment of the present invention;
FIG. 2 (a) is a schematic diagram of a meteorological variable box according to an embodiment of the present invention;
FIG. 2 (b) is a stacked histogram of correlation coefficients according to an embodiment of the present invention;
FIG. 3 is a graph showing the comparison of the normal gradient and the natural gradient calculation process according to the embodiment of the invention.
FIG. 4 is a schematic diagram of a Blending fusion step according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a wind power probability prediction model building method, which can predict and acquire uncertainty information of wind power, is beneficial to improving the operation reliability of a wind power grid and solves the problems of low wind power generation consumption level and difficult scheduling plan formulation caused by the lack of uncertainty information of wind power.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Fig. 1 is an initial data diagram of an embodiment of the present invention, as shown in fig. 1, all data are actually measured by a SCADA system in 2019 of a certain wind farm in northeast China. The data sampling interval is 15min, and 960 samples of 1 month, 1 day to 1 month and 10 days are taken to form an initial data set. Wherein: the input meteorological variables comprise wind direction, temperature, humidity, air pressure and wind speed of a wind farm collection point, and the output variables are wind power. As can be seen from fig. 1: the wind power and related meteorological variables have strong randomness and volatility, and probability prediction of the wind power is performed to extract related uncertainty information.
The wind power probability prediction model establishment method provided by the invention comprises the following steps:
s1, data preprocessing: removing abnormal values in the initial data set, and selecting weather variables with the relevance to the wind power greater than a preset threshold value as a training data set of a wind power probability prediction model based on a gray relevance theory;
s2, establishing an improved natural gradient lifting meta-model: based on a training data set, predicting parameter vectors of wind power probability distribution, establishing a connection between a general gradient and a natural gradient through Fisher information quantity, selecting classification and regression trees as a base learner, and establishing an improved natural gradient lifting meta-model to realize parameter vector updating;
s3, blending model fusion: and carrying out Blending model fusion on the improved natural gradient lifting meta-models, establishing a new meta-model for training, and outputting a final prediction statistical parameter vector.
In the step S1, abnormal values in the initial data set are removed, and based on a gray correlation theory, a meteorological variable with a degree of correlation with wind power greater than a preset threshold value is selected as a training data set of a wind power probability prediction model, and the method specifically comprises the following steps:
s101, eliminating abnormal values in initial data by using a box graph;
s102, using wind power as a reference data column and related meteorological variables as a comparison data column, performing initialization processing on each sequence, calculating association coefficients based on a gray association theory to represent association degrees of two groups of sequences, and selecting meteorological variables with association degrees larger than a threshold value as a training data set of a prediction model.
As shown in fig. 2 (a), in step S101, the outlier in the initial data is removed by using a box graph, which specifically includes:
the outlier cutoff upper and lower limits are determined by equation (1):
wherein: min and max represent the upper limit and the lower limit of data interception; q (Q) 1 、Q 3 Respectively representing upper and lower quartiles; iqr=q 3 -Q 1 。
Wind power is related to meteorological variables such as temperature, wind speed and the like. However, this correlation varies from one wind farm to another, as affected by factors such as the geographic location of the wind farm and the local microclimate of the location. In order to improve the generalization capability and robustness of the model, the patent adopts gray correlation analysis to screen the characteristic meteorological variables of the input data, and the screening result is shown in fig. 2 (b). In step S102, wind power is taken as a reference data sequence, related meteorological variables are taken as a comparison data sequence, each sequence is initialized, association coefficients are calculated based on a gray association theory to represent association degrees of two groups of sequences, and meteorological variables with association degrees larger than a threshold value are selected as training data sets of a prediction model, and the method specifically comprises the following steps:
1) For each variable time sequenceThe column is normalized by taking the kth of the n meteorological variable sequences as a comparison sequence S k (t) the wind power sequence is a reference sequence S 0 (t) calculating the difference between the two as the absolute value sequence delta k (t) as shown in formula (2), wherein k E (1, n),
Δ k (t)＝|S k (t)-S 0 (t)| (2)
2) Calculating the correlation coefficient eta k (t)：
Wherein: min (·) and Max (·) represent minimum and maximum values of the sequence, and ρ is a resolution coefficient;
3) Solving the association degree gamma k ：
Wherein: t (T) n Is the sequence length;
4) Setting a threshold valueAnd selecting meteorological variables with the association degree larger than a threshold value to form a training data set.
In the step S2, based on the training data set, the parameter vector of the wind power probability distribution is predicted, a relationship is established between a general gradient and a natural gradient through Fisher information, a classification tree and a regression tree are selected as a base learner, an improved natural gradient lifting meta-model is established, and the parameter vector update is realized, which specifically comprises:
s201, let D include n D Samples, m features, i.e. d= { (x) i ，y i )}(x i ∈R m ，y i E R), where x i Characterizing the eigenvector of the ith sample, y i Representing the label value corresponding to the ith sample, i E (1, n) D )；
S202, y i Shannon information amount of (c)Build a scoring function S (θ, y) for the benchmark i )：
S(θ,y i )＝-log P θ (y i ) (8)
Wherein: p (P) θ (y i ) Is y i Probability values in the predictive probability distribution; θ is a parameter vector of the predictive probability distribution;
s203, let-log P θ (y i ) F (θ), taylor expansion is performed on f (θ+d'), and the third order and above remainder are discarded:
wherein: d' is theta edgeMoving an infinitely small step vector; />Representing a natural gradient;
converting the European space into a statistical manifold, and processing the formula (12) under the Riemann space:
wherein the calculation of the primary term can be simplified to:
the remainder is denoted as:
wherein: psi (θ), a Riemann metric of statistical manifold at θ, is used to characterize P θ (yi) the Fisher information amount brought by:
thereby realizing the calculation of natural gradient through general gradient
S204, selecting classification and regression trees as base learners, and generating a group of new base learners along the generalized natural gradient direction by each iteration of the base learners, thereby establishing an improved natural gradient lifting meta-model;
s205, updating a parameter vector theta of wind power probability distribution:
wherein: θ 0 As an initial parameter vector, α m Is a scale factor, beta is a unified learning rate, B m Is a unified representation of the base learner.
Scaling factor alpha m The setting of the method is to avoid the situation that the local approximation is far away from the current parameter position in the calculation process to cause training failure, and the specific numerical values are selected in a linear search mode. The learning rate β is usually 0.1 or 0.01.
Based on the step 1, the invention analyzes the application defect of the traditional Boosting method in the wind power probability prediction problem, provides a new natural gradient calculation method in a targeted way, establishes a relation between a general gradient and a natural gradient through Fisher information quantity, selects classification and regression trees as a base learner, and establishes a natural gradient lifting meta-model. Wherein, the common gradient application defect is analyzed:
when the traditional Boosting method solves the point prediction problem, the optimal approximation function F (x) i ) For the purpose ofThe target minimizes the expected value of the loss function L on the training set. Wherein L and F (x) i ) Can be expressed as:
wherein: subscript M denotes the mth gradient boost phase; weight gamma M The optimal step length is obtained through linear search;for a normal gradient, each iteration of the algorithm produces a new basis learner in that direction, minimizing the loss function, namely:
wherein: d is F (xi) edgeAn infinitely small step size of the movement.
Unlike point prediction is: probability prediction problem to solve y i Is a complete probability distribution P of (2) θ (y i ) Is the object. When P θ (y i ) Re-parameterization intoAt the same time, from θ to θ+d and +.>To->The calculated common gradient changes, and the moving gauge of the updated parameters in the distribution space cannot be truly reflectedLaw. Namely: the common gradient cannot keep unchanged for the repartitioning, and defects exist when the probability prediction problem is solved.
The invention lays a foundation for describing natural gradient and analogies the loss function in the point prediction problem by y i Establishes a scoring function S (theta, y) based on the shannon information amount i )。
S(θ,y i )＝-log P θ (y i ) (8)
Wherein: p (P) θ (y i ) Is y i Probability values in the predictive probability distribution; θ is a parameter vector of the predictive probability distribution.
Let Q be y i The expression (9) is always true.
The part on the right side and the left side of the inequality sign is the divergence D under the scoring function KL (Q||P) representing the difference between the predicted distribution and the actual distribution. Namely:
the natural gradient is a gradient determined according to KL divergence on the statistical manifold, is the fastest rising direction in the Riemann space, and has invariance to the re-parameterization. Generating a new base learner along the natural gradient direction at each iteration theta, and acquiring the maximum score improvement, namely:
wherein:representing a natural gradient; d' is theta edge->An infinitely small step vector of movement.
Because the related concepts of the natural gradient solution are taken from the information geometry, the method brings inconvenience to popularization and application in actual engineering. The invention focuses on improving the solving process of the natural gradient, establishes a relation between the general gradient and the natural gradient through Fisher information quantity, and calculates the natural gradient through the general gradient.
Unlike the point prediction model proposed by other inventions, the point prediction model is: aiming at wind power probability prediction, the model building method provided by the invention emphasizes a parameter vector theta for predicting wind power probability distribution, namely: uncertainty information of wind power can be predicted and obtained. The specific calculation process is as follows:
calculation first of all by θ 0 For initial parameter vectors, the essence is to fit y i Is arranged on the edge of the substrate; subsequently at the mth iteration, y is calculated i And corresponding parameter vector thereofNatural gradient +.>Fitting in this direction generates a new set of basis learners, thus enabling parameter vector updates.
The meta-model is fused, so that the learning effect can be enhanced, excessive redundancy of the whole model is not caused, and only part of the point prediction models in the current related field are fused by adopting a Stacking model. However, stacking model fusion is too complex, and the problem of data crossing of training data referencing global statistics occurs in the calculation process, so that the method is not suitable for solving the problem of wind power probability prediction.
In order to overcome the defects, the invention is based on the simple fusion of the Blending model, overcomes the advantage of data crossing, and fuses a plurality of improved natural gradient lifting meta-models, wherein the fusion step is shown in figure 4. In the step S3, a Blending model is performed on the plurality of improved natural gradient lifting meta-models, a new meta-model is built for training, so as to output a final predicted statistical parameter vector, which specifically includes:
s301, initial dataset segmentation: dividing an original training data set into a sub training set DT and a test set DA according to a proportion, and defining an original prediction data set as DP;
s302, model fusion:
given the confidence level, building V NGBoost meta-models MO 1 、MO 2 、…、MO V Learning DT by using the metamodels, and outputting a predicted result DA_ P, DP _P of DA and DP on the metamodels after training is completed, wherein DA_ P, DP _P is an initial statistical parameter vector of the predicted values corresponding to DA and DP;
the predicted mean value determined by DA_P and the actual result DA_OUT corresponding to the original DA data form a new data set, and a new meta-model MO is established DA Training and obtaining predicted output MO DA P, where MO is DA P is the corrected predicted statistical parameter vector; compared with DA_P, MODA_P has higher accuracy and smaller sharpness, and the advantages of model fusion are reflected;
MO is prepared from DA P and DP P form new data set, and new meta-model MO is built P Training is carried out, so that a final prediction statistical parameter vector is output, the upper limit and the lower limit of a predicted value under a given confidence level are calculated through the vector, and the points are connected into a predicted value upper limit curve and a predicted value lower limit curve.
According to the wind power probability prediction model establishment method provided by the invention, a natural gradient lifting meta-model of generalized natural gradient calculation is utilized; compared with the traditional point prediction model, the model provided by the invention solves the application defect of the traditional Boosting algorithm in solving the problem of wind power probability prediction, has higher generalization and robustness, has higher prediction area coverage rate and smaller average width occupation ratio of a prediction interval, has the reinforced prediction effect, and simultaneously has lower model redundancy, and the model building method is more suitable for practical engineering application, thereby being beneficial to improving the power grid scheduling economy and the running safety of a wind power plant with renewable energy grid connection.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (1)
1. The wind power probability prediction model building method is characterized by comprising the following steps of:
s1, data preprocessing: removing abnormal values in the initial data set, and selecting weather variables with the relevance to the wind power greater than a preset threshold value as a training data set of a wind power probability prediction model based on a gray relevance theory;
s2, establishing an improved natural gradient lifting meta-model: based on a training data set, predicting parameter vectors of wind power probability distribution, establishing a connection between a general gradient and a natural gradient through Fisher information quantity, selecting classification and regression trees as a base learner, and establishing an improved natural gradient lifting meta-model to realize parameter vector updating;
s3, blending model fusion: performing Blending model fusion on a plurality of improved natural gradient lifting meta-models, establishing a new meta-model for training, and outputting a final prediction statistical parameter vector;
in the step S1, abnormal values in the initial data set are removed, and based on a gray correlation theory, a meteorological variable with a degree of correlation with wind power greater than a preset threshold value is selected as a training data set of a wind power probability prediction model, and the method specifically comprises the following steps:
s101, eliminating abnormal values in initial data by using a box graph;
s102, carrying out initial value processing on each sequence by taking wind power as a reference data sequence and relevant meteorological variables as a comparison data sequence, calculating association coefficients based on a gray association theory to represent the association degree of two groups of sequences, and selecting meteorological variables with the association degree larger than a threshold value as a training data set of a prediction model;
in the step S101, the step of eliminating the outlier in the initial data by using the box graph specifically includes:
the outlier cutoff upper and lower limits are determined by equation (1):
wherein: min and max represent the upper limit and the lower limit of data interception; q (Q) 1 、Q 3 Respectively representing upper and lower quartiles; iqr=q 3 -Q 1 ；
The wind power is taken as a reference data column, related meteorological variables are taken as a comparison data column, each sequence is subjected to initial value processing, the association coefficient is calculated based on a gray association theory to represent the association degree of two groups of sequences, and the meteorological variables with the association degree larger than a threshold value are selected as training data sets of a prediction model, and the method specifically comprises the following steps:
1) Normalizing the time series of each variable, and taking the kth of the n meteorological variable series as a comparison series S k (t) the wind power sequence is a reference sequence S 0 (t) calculating the difference between the two as the absolute value sequence delta k (t) as shown in formula (2), wherein k E (1, n),
Δ k (t)＝|S k (t)-S 0 (t)| (2)
2) Calculating the correlation coefficient eta k (t)：
Wherein: min (·) and Max (·) represent minimum and maximum values of the sequence, and ρ is a resolution coefficient;
3) Solving the association degree gamma k ：
Wherein: t (T) n Is the sequence length;
4) Setting a threshold valueSelecting meteorological variables with the association degree larger than a threshold value to form a training data set;
in the step S2, based on the training data set, the parameter vector of the wind power probability distribution is predicted, a relationship is established between a general gradient and a natural gradient through Fisher information, a classification tree and a regression tree are selected as a base learner, an improved natural gradient lifting meta-model is established, and the parameter vector update is realized, which specifically comprises:
s201, let D include n D Samples, m features, i.e. d= { (x) i ，y i )}(x i ∈R m ，y i E R), where x i Characterizing the eigenvector of the ith sample, y i Representing the label value corresponding to the ith sample, i E (1, n) D )；
S202, y i Establishes a scoring function S (theta, y) based on the shannon information amount i )：
S(θ,y i )＝-logP θ (y i ) (8)
Wherein: p (P) θ (y i ) Is y i Probability values in the predictive probability distribution; θ is a parameter vector of the predictive probability distribution;
s203, make-log P θ (y i ) F (θ), taylor expansion is performed on f (θ+d'), and the third order and above remainder are discarded:
wherein: d' is theta edgeMoving an infinitely small step vector; />Representing a natural gradient;
converting the European space into a statistical manifold, and processing the formula (12) under the Riemann space:
wherein the calculation of the primary term can be simplified to:
the remainder is denoted as:
wherein: psi (θ), a Riemann metric of statistical manifold at θ, is used to characterize P θ (yi) the Fisher information amount brought by:
thereby realizing the calculation of natural gradient through general gradient
S204, selecting classification and regression trees as base learners, and generating a group of new base learners along the generalized natural gradient direction by each iteration of the base learners, thereby establishing an improved natural gradient lifting meta-model;
s205, updating a parameter vector theta of wind power probability distribution:
wherein: θ 0 As an initial parameter vector, α m Is a scale factor, beta is a unified learning rate, B m A unified representation of the base learner;
in the step S3, a Blending model is performed on the plurality of improved natural gradient lifting meta-models, a new meta-model is built for training, so as to output a final predicted statistical parameter vector, which specifically includes:
s301, initial dataset segmentation: dividing an original training data set into a sub training set DT and a test set DA according to a proportion, and defining an original prediction data set as DP;
s302, model fusion:
given the confidence level, building V NGBoost meta-models MO 1 、MO 2 、…、MO V Learning DT by using the metamodels, and outputting a predicted result DA_ P, DP _P of DA and DP on the metamodels after training is completed, wherein DA_ P, DP _P is an initial statistical parameter vector of the predicted values corresponding to DA and DP;
the predicted mean value determined by DA_P and the actual result DA_OUT corresponding to the original DA data form a new data set, and a new meta-model MO is established DA Training and obtaining predicted output MO DA P, where MO is DA P is the corrected predicted statistical parameter vector;
MO is prepared from DA P and DP P form new data set, and new meta-model MO is built P Training is carried out, so that a final prediction statistical parameter vector is output, the upper limit and the lower limit of a predicted value under a given confidence level are calculated through the vector, and the points are connected into a predicted value upper limit curve and a predicted value lower limit curve.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010834929.XA CN111985711B (en) | 2020-08-19 | 2020-08-19 | Wind power probability prediction model building method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010834929.XA CN111985711B (en) | 2020-08-19 | 2020-08-19 | Wind power probability prediction model building method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111985711A CN111985711A (en) | 2020-11-24 |
| CN111985711B true CN111985711B (en) | 2024-02-02 |
Family
ID=73434116
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010834929.XA Active CN111985711B (en) | 2020-08-19 | 2020-08-19 | Wind power probability prediction model building method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111985711B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529719A (en) * | 2016-11-08 | 2017-03-22 | 中国地质大学（武汉） | Method of predicting wind power of wind speed fusion based on particle swarm optimization algorithm |
| CN107221933A (en) * | 2016-03-22 | 2017-09-29 | 中国电力科学研究院 | A kind of probability load flow calculation method |
| CN109214566A (en) * | 2018-08-30 | 2019-01-15 | 华北水利水电大学 | Short-term wind power prediction method based on shot and long term memory network |
| CN110264012A (en) * | 2019-06-26 | 2019-09-20 | 山东大学 | Renewable energy power combination prediction technique and system based on empirical mode decomposition |
| CN110334875A (en) * | 2019-07-10 | 2019-10-15 | 东北电力大学 | Meter and the wind power combined probability prediction technique of evaluation index conflict |
| CN111191857A (en) * | 2020-02-18 | 2020-05-22 | 上海积成能源科技有限公司 | Wind power generation prediction system model based on gradient lifting algorithm |
| CN111489038A (en) * | 2020-04-14 | 2020-08-04 | 宁波金风绿能能源有限公司 | Method for improving accuracy of ultra-short-term power prediction of wind power plant |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103927695B (en) * | 2014-04-22 | 2017-11-24 | 国家电网公司 | Ultrashort-term wind power prediction method based on self study complex data source |
-
2020
- 2020-08-19 CN CN202010834929.XA patent/CN111985711B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107221933A (en) * | 2016-03-22 | 2017-09-29 | 中国电力科学研究院 | A kind of probability load flow calculation method |
| CN106529719A (en) * | 2016-11-08 | 2017-03-22 | 中国地质大学（武汉） | Method of predicting wind power of wind speed fusion based on particle swarm optimization algorithm |
| CN109214566A (en) * | 2018-08-30 | 2019-01-15 | 华北水利水电大学 | Short-term wind power prediction method based on shot and long term memory network |
| CN110264012A (en) * | 2019-06-26 | 2019-09-20 | 山东大学 | Renewable energy power combination prediction technique and system based on empirical mode decomposition |
| CN110334875A (en) * | 2019-07-10 | 2019-10-15 | 东北电力大学 | Meter and the wind power combined probability prediction technique of evaluation index conflict |
| CN111191857A (en) * | 2020-02-18 | 2020-05-22 | 上海积成能源科技有限公司 | Wind power generation prediction system model based on gradient lifting algorithm |
| CN111489038A (en) * | 2020-04-14 | 2020-08-04 | 宁波金风绿能能源有限公司 | Method for improving accuracy of ultra-short-term power prediction of wind power plant |
Non-Patent Citations (4)
| Title |
|---|
| "A data-driven multi-model methodology with deep feature selection for short-term wind forecasting";Cong Feng 等;《Applied Energy》;第190卷;1245-1257 * |
| "基于提升回归树与随机森林的风电功率集成预测方法";李国 等;《电力系统及其自动化学报》(第11期);74-78 * |
| "用于含风电场的电力系统概率潮流计算的高斯混合模型";叶林 等;《中国电机工程学报》;第37卷(第15期);4379-4387，4578 * |
| "风电功率预测方法综述及发展研究";牛东晓 等;《现代电力》;第30卷(第4期);24-28 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111985711A (en) | 2020-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109802430B (en) | Wind power grid control method based on LSTM-Attention network | |
| Xiaoyun et al. | Short-term prediction of wind power based on deep long short-term memory | |
| CN111008728B (en) | Prediction method for short-term output of distributed photovoltaic power generation system | |
| CN103117546B (en) | A kind of Ultrashort-term slide prediction method for wind power | |
| CN113657662B (en) | Downscaling wind power prediction method based on data fusion | |
| CN110766200A (en) | Method for predicting generating power of wind turbine generator based on K-means mean clustering | |
| CN113379142B (en) | Short-term wind power prediction method based on wind speed correction and fusion model | |
| CN110601250B (en) | Method for evaluating stability of static voltage of access point based on new energy power station | |
| CN115345076B (en) | Wind speed correction processing method and device | |
| CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
| Chen et al. | Research on wind power prediction method based on convolutional neural network and genetic algorithm | |
| CN113822418A (en) | Wind power plant power prediction method, system, device and storage medium | |
| Li et al. | Deep spatio-temporal wind power forecasting | |
| CN115049163A (en) | Photovoltaic region prediction method and device based on tabu search and deep stable learning | |
| CN111985711B (en) | Wind power probability prediction model building method | |
| CN116565840A (en) | High-precision wind speed soft measurement method for wind power prediction of wind power plant | |
| CN116663393A (en) | Random forest-based power distribution network continuous high-temperature fault risk level prediction method | |
| CN116231749A (en) | New energy power system dispatching method based on digital twin | |
| CN116299774A (en) | Artificial intelligence numerical weather forecast method and system | |
| CN113779892B (en) | Method for predicting wind speed and wind direction | |
| CN115912334A (en) | Method for establishing prediction model of output guarantee rate of wind power plant and prediction method | |
| CN113095547B (en) | Short-term wind power prediction method based on GRA-LSTM-ICE model | |
| CN111177973B (en) | Photovoltaic array online modeling method based on reinforcement learning | |
| CN114298441A (en) | Photovoltaic power prediction method and system | |
| CN117688367B (en) | Wind power generation ultra-short term power prediction method and device based on instant learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |