CN112070160B - Multilayer collaborative real-time classification early warning method for wind power climbing event - Google Patents

Multilayer collaborative real-time classification early warning method for wind power climbing event Download PDF

Info

Publication number
CN112070160B
CN112070160B CN202010934286.6A CN202010934286A CN112070160B CN 112070160 B CN112070160 B CN 112070160B CN 202010934286 A CN202010934286 A CN 202010934286A CN 112070160 B CN112070160 B CN 112070160B
Authority
CN
China
Prior art keywords
wind power
time point
time
climbing
value
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
Application number
CN202010934286.6A
Other languages
Chinese (zh)
Other versions
CN112070160A (en
Inventor
何耀耀
王云
肖经凌
张婉莹
陈悦
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.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
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 Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202010934286.6A priority Critical patent/CN112070160B/en
Publication of CN112070160A publication Critical patent/CN112070160A/en
Application granted granted Critical
Publication of CN112070160B publication Critical patent/CN112070160B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Marketing (AREA)
  • General Engineering & Computer Science (AREA)
  • Water Supply & Treatment (AREA)
  • Computational Linguistics (AREA)
  • Public Health (AREA)
  • Primary Health Care (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (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 multilayer collaborative real-time classification early warning method for a wind power climbing event, which comprises the following steps: 1, acquiring historical wind power data for preprocessing; 2, formulating a wind power climbing event classification strategy; 3, establishing a multi-layer collaborative prediction model of a decomposition layer, a prediction layer, a correction layer and a feedback layer; and 4, identifying wind power climbing events of different categories to perform real-time early warning. The method utilizes EMD decomposition, GRU prediction, SVR correction and multistep rolling prediction of actual measurement information feedback, and carries out real-time early warning after identifying different types of wind power climbing events according to the classification criteria of the wind power climbing events, and can take corresponding measures in time before the climbing events occur, thereby ensuring the safe and stable operation of the power system.

Description

Multilayer collaborative real-time classification early warning method for wind power climbing event
Technical Field
The invention belongs to the technical field of prediction and control of electric power systems, and particularly relates to a multilayer collaborative real-time classification early warning method for wind power climbing events.
Background
In order to cope with the shortage of the conventional fossil energy and the global severe environmental problem, the ratio of new energy to the power grid is continuously increasing. Wind energy is a green and clean renewable energy source, and is rapidly developed and widely applied in the global scope. Along with the improvement of wind power permeability, the wind power is greatly changed in a short time under the influence of extreme weather, and a wind power climbing event is easily generated, so that a large amount of power shortage in a short time of a system is caused, peak load regulation and frequency modulation are influenced, and a large amount of load loss is caused. Therefore, accurate prediction and quick early warning of the wind power climbing event have important significance for ensuring the safety, stability, economy and reliability of the power system.
The climbing event prediction method is mainly divided into an indirect prediction method and a direct prediction method. The method is widely applied to indirect prediction methods, namely a climbing event is identified based on wind power prediction information, however, when the wind power prediction is carried out, in order to improve the overall prediction precision, the method often ignores terminal data, so that climbing information is lost, and the probability of correctly predicting the climbing event is low. Therefore, the method for predicting the climbing event with consideration of the wind power prediction accuracy and the climbing information integrity becomes a technical problem to be solved at present.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a multilayer collaborative real-time classification early warning method for wind power climbing events, so that wind power prediction information can be fully utilized to accurately early warn different types of climbing events, corresponding defense measures can be taken in time for various climbing events, wind power friendly grid connection is promoted, the occurrence of wind abandonment is reduced, and the safe, stable, economic and reliable operation of a power system is ensured.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention relates to a multilayer collaborative real-time classification early warning method for wind power climbing events, which is characterized by comprising the following steps of:
s1) acquiring historical data of a wind power time sequence for preprocessing to obtain a preprocessed wind power time sequence, and recording the preprocessed wind power time sequence as (Y (t) } t＝1,2,...,T Y (T) is the wind power at the time point T, and T is the total number of samples;
s2) formulating a wind power climbing event classification strategy:
s2.1) wind power sequence (Y (t + M) of M time points in the future after the t time point m＝1,2,...,M According to the formula (1), establishing a wind power climbing trend factor R of a t time point t ：
Figure BDA0002671385520000011
In the formula (1), Y (t + m) is the wind power of the mth time point in the future after the t time point; sign (.) is a sign function, R t The maximum value is M, and the minimum value is-M;
s2.2) setting the climbing threshold value to be delta, and defining the wind power climbing amplitude W of the t time point according to the formula (2) t ：
W t ＝max({Y(t+m)} m＝1,2,...,M )-min({Y(t+m)} m＝1,2,...,M ) (2)
S2.3) combining the wind power climbing trend factor R t A climbing threshold value delta and a climbing amplitude W t And formulating four classifications and classification criteria of an uphill event, a downhill event, a strong wave event and a non-uphill event of the wind power:
1) Wind power climbing trend factor R at t-time point t If = M, identifying an uphill event and a non-uphill event according to equation (3);
Figure BDA0002671385520000021
2) Wind power climbing trend factor R at t time point t If the mark is "= -M, identifying a climbing event and a non-climbing event according to the formula (4);
Figure BDA0002671385520000022
3) Wind power climbing trend factor R at t-time point t When the condition belongs to (-M, M), identifying a strong fluctuation event and a non-climbing event according to a formula (5);
Figure BDA0002671385520000023
s3) establishing a multi-layer collaborative prediction model of a decomposition layer, a prediction layer, a correction layer and a feedback layer of M time points in the future after T time point:
s3.1) defining the training sample set as
Figure BDA0002671385520000024
And initialize
Figure BDA0002671385520000025
Defining the number of training samples as
Figure BDA00026713855200000210
And initialize
Figure BDA0002671385520000026
S3.2) establishing a decomposition layer:
set of training samples using EMD method
Figure BDA0002671385520000027
Decomposing to obtain K intrinsic mode components
Figure BDA0002671385520000028
And a residual component
Figure BDA0002671385520000029
Wherein c is k (t) represents the value of the kth eigenmode component at the time point t, and r (t) represents the value of the residual component at the time point t;
calculating the energy entropy { H ] of each eigenmode component and residual component k } k＝1,2,...,K+1 Combining the components of similar energy entropy to obtain S subsequences
Figure BDA0002671385520000031
S＜K，H k Is the energy entropy, C, of the kth component s (t) is the value of the s-th subsequence at time point t;
s3.3) establishing a prediction layer:
s3.3.1) for the s-th sub-sequence
Figure BDA0002671385520000032
Historical data { C) of P time points before t time point by using s-th subsequence s (t-p)} p＝1,2,...,P Rolling the value C of the predicted t time point s (t) wherein C s (t-P) is the data of the P-th time point before the time point t, so as to construct P interpretation variables { C s (t-p)} p＝1,2,...,P And a response variable C s (t)，
Figure BDA0002671385520000033
In all
Figure BDA0002671385520000034
Training samples;
s3.3.2) all explanatory variables
Figure BDA0002671385520000035
And all response variables
Figure BDA0002671385520000036
Respectively as the input and output of the gate control cycle unit neural network GRU prediction model, and training to obtain the trained GRU prediction model and corresponding fitting value
Figure BDA0002671385520000037
Figure BDA0002671385520000038
Fitting values of points at the t moment for the s-th subsequence;
s3.3.3) rolling the data backwards by a time point, namely
Figure BDA0002671385520000039
As input to the trained GRU prediction model, thereby obtaining
Figure BDA00026713855200000310
Predicted value of time point
Figure BDA00026713855200000311
S3.3.4) assigning S +1 to S, judging whether S > S is true, if so, finishing the S subsequencesPredicting to obtain S subsequences from P +1 to
Figure BDA00026713855200000322
Fitting value of time GRU prediction model
Figure BDA00026713855200000312
And in
Figure BDA00026713855200000313
Predicted value of time point
Figure BDA00026713855200000314
Otherwise, returning to the step S3.3.1;
s3.3.5) respectively solving the wind power from the moment P +1 to the moment
Figure BDA00026713855200000323
Integrated fitting value of time of day
Figure BDA00026713855200000315
And
Figure BDA00026713855200000316
initial predicted value of time point
Figure BDA00026713855200000317
Figure BDA00026713855200000318
Figure BDA00026713855200000319
S3.4) establishing a correction layer:
s3.4.1) extracting residual sequence of GRU prediction model according to formula (9)
Figure BDA00026713855200000320
Figure BDA00026713855200000321
In the formula (9), e (t) is the value of the residual sequence at the time t;
s3.4.2) utilizing residual error data { e (t-P) }of P time points before t time points of the residual error sequence p＝1,2,...,P Predicting data e (t) at the t moment, wherein e (t-p) is the data of the p-th moment point before the t moment point,
Figure BDA0002671385520000041
co-construction
Figure BDA0002671385520000042
Training samples;
s3.4.3) will
Figure BDA0002671385520000043
And
Figure BDA0002671385520000044
respectively serving as the input and the output of the SVR correction model and training to obtain a trained SVR correction model;
s3.4.4) rolling the data backwards by a time point, namely
Figure BDA0002671385520000045
As input of the trained SVR correction model, thereby obtaining
Figure BDA0002671385520000046
Residual prediction value of time point
Figure BDA0002671385520000047
S3.4.5) prediction layer Using equation (10)
Figure BDA0002671385520000048
Wind power initial prediction result of time point
Figure BDA0002671385520000049
Performing correction to obtain the final product
Figure BDA00026713855200000410
Final predicted value of time point
Figure BDA00026713855200000411
Figure BDA00026713855200000412
S3.5) establishing a feedback layer:
by wind-electric plants
Figure BDA00026713855200000413
Actual measurement information of time point
Figure BDA00026713855200000414
Replacing the final predicted value at the corresponding time
Figure BDA00026713855200000415
And incorporating the training samples into the training samples to obtain an updated training sample set according to equations (11) and (12)
Figure BDA00026713855200000416
And updated total number of training samples
Figure BDA00026713855200000417
Figure BDA00026713855200000418
Figure BDA00026713855200000419
S3.6) will
Figure BDA00026713855200000420
Is assigned to
Figure BDA00026713855200000421
Will be provided with
Figure BDA00026713855200000422
Is assigned to
Figure BDA00026713855200000425
And then, repeating the step S3.2-the step S3.5 until the prediction of the future M wind power values after the T moment is finished to obtain M wind power predicted values
Figure BDA00026713855200000423
S4) combining the classification criterion of the wind power climbing event in the step S2 and the predicted value of the wind power at the future M moment predicted by the multi-layer collaborative prediction model in the step S3
Figure BDA00026713855200000424
And identifying the wind power climbing event, thereby realizing the real-time early warning of various wind power climbing events.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the wind power grade climbing event classification method, based on the multistep rolling prediction information of the wind power, a classification strategy of the wind power grade climbing event is formulated, historical data of the wind power and data at future moments are fully utilized, and the false alarm rate of the wind power grade climbing event is effectively reduced;
2. the method has the advantages that the method utilizes EMD decomposition to decompose the wind power into a series of simple and stable subsequences, reduces the prediction difficulty and effectively improves the prediction accuracy;
3. according to the method, error correction is added on the basis of primary prediction of the wind power, secondary prediction is performed on information which is possibly lost in the prediction process, and meanwhile, real-time feedback is performed by using the measured information of the wind power plant, so that the loss of the information of the climbing event in the wind power data is avoided, the accuracy of early warning of the climbing event of the wind power is greatly improved, and powerful support is provided for safe, stable and economic operation of the system.
Drawings
FIG. 1 is an overall flow diagram of the present invention;
FIG. 2 is a block diagram of a gated cyclic unit neural network of the present invention.
Detailed Description
The technical solution of this patent will be further described in detail with reference to specific embodiments.
As shown in fig. 1, a multilayer collaborative real-time classification early warning method for a wind power climbing event includes the following steps:
s1) acquiring historical data of a wind power time sequence for preprocessing to obtain a preprocessed wind power time sequence, and recording the preprocessed wind power time sequence as (Y (t) } t＝1,2,...,T Y (T) is the wind power at the time point T, and T is the total number of samples;
s2) formulating a wind power climbing event classification strategy:
s2.1) wind power sequence (Y (t + M) of M time points in the future after the t time point m＝1,2,...,M According to the formula (1), establishing a wind power climbing trend factor R of a t time point t ：
Figure BDA0002671385520000051
In the formula (1), Y (t + m) is the wind power of the mth time point in the future after the t time point; sign () is a sign function and satisfies:
Figure BDA0002671385520000052
as defined in the meaning of the symbol, R t The maximum value is M, and the minimum value is-M;
s2.2) setting a climbing threshold value delta, and defining the wind power climbing amplitude W of a t time point according to the formula (3) t ：
W t ＝max({Y(t+m)} m＝1,2,...,M )-min({Y(t+m)} m＝1,2,...,M ) (3)
S2.3) combining wind power climbing trend factor R t A climbing threshold value delta and a climbing amplitude W t And formulating four classifications and classification criteria of an ascending event, a descending event, a strong fluctuation event and a non-climbing event of the wind power:
1) Wind power climbing trend factor R at t time point t When = M, the wind power is continuously increased in the range of P time points in the future after t time, and an ascending event and a non-ascending event are identified according to the formula (4);
Figure BDA0002671385520000061
2) Wind power climbing trend factor R at t-time point t When the power is not equal to M, the wind power is continuously reduced in the range of P time points in the future after t time, and a climbing event and a non-climbing event are identified according to the formula (5);
Figure BDA0002671385520000062
3) Wind power climbing trend factor R at t time point t When the wind power belongs to (-M, M), expressing that the wind power changes non-monotonously in the range of P time points in the future after t time, and identifying a strong fluctuation event and a non-climbing event according to a formula (6);
Figure BDA0002671385520000063
s3) establishing a multi-layer collaborative prediction model of a decomposition layer, a prediction layer, a correction layer and a feedback layer of future M time points after T time points:
s3.1) defining the training sample set as
Figure BDA0002671385520000064
And initialize
Figure BDA0002671385520000065
Defining the number of training samples as
Figure BDA00026713855200000611
And initialize
Figure BDA0002671385520000066
S3.2) establishing a decomposition layer:
set of training samples using EMD method
Figure BDA0002671385520000067
Decomposing to obtain K intrinsic mode components
Figure BDA0002671385520000068
And a residual component
Figure BDA0002671385520000069
And satisfies:
Figure BDA00026713855200000610
in the formula (7), c k (t) represents the value of the kth eigenmode component at the time point t, and r (t) represents the value of the residual component at the time point t;
the energy entropy { H } of each eigenmode component and residual component is calculated from equation (8) k } k＝1,2,...,K+1 ：
Figure BDA0002671385520000071
In formula (8), E k Is the energy value of the kth component; p k Represents the proportion of the energy of the kth component in the total energy, H k Is the energy entropy of the kth component;
combining the components with similar energy entropy to obtain S subsequences
Figure BDA0002671385520000072
S＜K，C s (t) is the value of the s-th subsequence at time point t;
s3.3) establishing a prediction layer:
s3.3.1) for the s-th sub-sequence
Figure BDA0002671385520000073
Historical data { C) of P time points before t time point by using s-th subsequence s (t-p)} p＝1,2,...,P Rolling the value C of the predicted t time point s (t) wherein C s (t-P) is the data of the P-th time point before the time point t, so as to construct P interpretation variables { C s (t-p)} p＝1,2,...,P And a response variable C s (t)，
Figure BDA0002671385520000074
In all
Figure BDA0002671385520000075
Training samples;
s3.3.2) As shown in FIG. 2, the gated recurrent unit neural network GRU prediction model is a variant of the recurrent neural network, "+" denotes addition, "×" denotes the Hadamard product of the matrix, σ and tanh denote the sigmoid function and tanh activation function, respectively, x t Is the input quantity at time t, h' t For the memory content at time t, h t Is the final memory value at time t, y t Is the output value at time t, z t And r t For the update gate and the reset gate at time t, a gated cyclic unit neural network GRU prediction model can be represented by equation (9):
Figure BDA0002671385520000076
in the formula (9), ω is z For updating the weight matrix of the gate input, ω r A weight matrix which is the input quantity of the reset gate; mu is a weight matrix from an input layer to a hidden layer, omega is a self-feedback weight vector of the hidden layer, and upsilon isA weight matrix from the hidden layer to the output layer;
all the explained variables
Figure BDA0002671385520000077
And all response variables
Figure BDA0002671385520000078
Respectively as the input and output of the gate control cycle unit neural network GRU prediction model, and training to obtain the trained GRU prediction model and corresponding fitting value
Figure BDA0002671385520000079
Figure BDA00026713855200000710
Fitting values of the s-th subsequence at a time point t;
s3.3.3) rolling the data backwards by a time point, namely
Figure BDA0002671385520000081
As input of the trained GRU prediction model, thereby obtaining
Figure BDA0002671385520000082
Predicted value of time point
Figure BDA0002671385520000083
S3.3.4) assigning S +1 to S, then judging whether S > S is true, if so, completing the prediction of S subsequences, thereby obtaining S subsequences from P +1 to
Figure BDA00026713855200000825
Fitting value of time GRU prediction model
Figure BDA0002671385520000084
And in
Figure BDA0002671385520000085
Predicted value of time point
Figure BDA0002671385520000086
Otherwise, returning to the step S3.3.1;
s3.3.5) respectively calculating the wind power from the moment P +1 to the moment P +1 by using the formula (10) and the formula (11)
Figure BDA00026713855200000826
Integrated fitting value of time of day
Figure BDA0002671385520000087
And
Figure BDA0002671385520000088
initial predicted value of time point
Figure BDA0002671385520000089
Figure BDA00026713855200000810
Figure BDA00026713855200000811
S3.4) establishing a correction layer:
s3.4.1) extracting residual sequence of GRU prediction model according to formula (12)
Figure BDA00026713855200000812
Figure BDA00026713855200000813
In the formula (9), e (t) is the value of the residual sequence at the time t;
s3.4.2) utilizing residual error data { e (t-P) }of P time points before t time points of the residual error sequence p＝1,2,...,P Predicting data e (t) at the t moment, wherein e (t-p) is the data of the p-th moment point before the t moment point,
Figure BDA00026713855200000814
co-construction
Figure BDA00026713855200000815
Training samples;
s3.4.3) will
Figure BDA00026713855200000816
And
Figure BDA00026713855200000817
respectively serving as the input and the output of the SVR correction model and training to obtain a trained SVR correction model;
s3.4.4) rolling the data backwards by a time point, namely
Figure BDA00026713855200000818
As input of the trained SVR correction model, thereby obtaining
Figure BDA00026713855200000819
Residual prediction value of time point
Figure BDA00026713855200000820
S3.4.5) prediction layer Using equation (13)
Figure BDA00026713855200000827
Wind power initial prediction result of time point
Figure BDA00026713855200000821
Correcting to obtain the final product
Figure BDA00026713855200000822
Final predicted value of time point
Figure BDA00026713855200000823
Figure BDA00026713855200000824
S3.5) establishing a feedback layer:
by wind-electric plants
Figure BDA0002671385520000091
Measured information of a time point
Figure BDA0002671385520000092
Replacing the final predicted value at the corresponding time
Figure BDA0002671385520000093
And incorporating the training samples into the training samples to obtain an updated training sample set according to equations (14) and (15)
Figure BDA0002671385520000094
And updated total number of training samples
Figure BDA0002671385520000095
Figure BDA0002671385520000096
Figure BDA0002671385520000097
S3.6) will
Figure BDA0002671385520000098
Assign to
Figure BDA0002671385520000099
Will be provided with
Figure BDA00026713855200000910
Assign to
Figure BDA00026713855200000911
Thereafter, step S3.2-step S3.5 are repeatedUntil the prediction of the future M wind power values after T moment is finished, obtaining M wind power predicted values
Figure BDA00026713855200000912
S4) combining the classification criterion of the wind power climbing event in the step S2 and the predicted value of the wind power at the future M moment predicted by the multi-layer collaborative prediction model in the step S3
Figure BDA00026713855200000913
And identifying the wind power climbing event, thereby realizing the real-time early warning of various wind power climbing events.

Claims (1)

1. A multilayer collaborative real-time classification early warning method for wind power climbing events is characterized by comprising the following steps:
s1) acquiring historical data of a wind power time sequence for preprocessing to obtain a preprocessed wind power time sequence, and recording the preprocessed wind power time sequence as (Y (t) } t＝1,2,...,T Y (T) is the wind power at the time point T, and T is the total number of samples;
s2) formulating a wind power climbing event classification strategy:
s2.1) wind power sequence (Y (t + M) of M time points in the future after the t time point m＝1,2,...,M According to the formula (1), establishing a wind power climbing trend factor R of a t time point t ：
Figure FDA0003855166150000011
In the formula (1), Y (t + m) is the wind power of the mth time point in the future after the t time point; sign (. Quadrature.) is a sign function, R t The maximum value is M, and the minimum value is-M;
s2.2) setting the climbing threshold value to be delta, and defining the wind power climbing amplitude W of the t time point according to the formula (2) t ：
W t ＝max({Y(t+m)} m＝1,2,...,M )-min({Y(t+m)} m＝1,2,...,M ) (2)
S2.3) combining the wind power climbing trend factor R t A climbing threshold value delta and a climbing amplitude W t And formulating four classifications and classification criteria of an ascending event, a descending event, a strong fluctuation event and a non-climbing event of the wind power:
1) Wind power climbing trend factor R at t time point t If = M, identifying an uphill event and a non-uphill event according to equation (3);
Figure FDA0003855166150000012
2) Wind power climbing trend factor R at t time point t If the value is = M, identifying a climbing event and a non-climbing event according to the formula (4);
Figure FDA0003855166150000013
3) Wind power climbing trend factor R at t time point t When the element belongs to (-M, M), identifying a strong fluctuation event and a non-climbing event according to a formula (5);
Figure FDA0003855166150000014
s3) establishing a multi-layer collaborative prediction model of a decomposition layer, a prediction layer, a correction layer and a feedback layer of future M time points after T time points:
s3.1) defining the training sample set as
Figure FDA00038551661500000227
And initialize
Figure FDA0003855166150000021
Defining the number of training samples as
Figure FDA0003855166150000022
And initialize
Figure FDA0003855166150000023
S3.2) establishing a decomposition layer:
set of training samples using EMD method
Figure FDA00038551661500000228
Decomposing to obtain K intrinsic mode components
Figure FDA0003855166150000024
And a residual component
Figure FDA0003855166150000025
Wherein c is k (t) represents the value of the kth eigenmode component at the time point t, and r (t) represents the value of the residual component at the time point t;
calculating the energy entropy { H ] of each eigenmode component and residual component k } k＝1,2,...,K+1 And combining the components of similar energy entropy to obtain S subsequences
Figure FDA0003855166150000026
S＜K，H k Is the energy entropy of the k-th component, C s (t) is the value of the s-th subsequence at time point t;
s3.3) establishing a prediction layer:
s3.3.1) for the s-th sub-sequence
Figure FDA0003855166150000027
Utilizing historical data of P time points before t time point of the s-th subsequence
Figure FDA0003855166150000028
Rolling the value C of the predicted t time point s (t) wherein C s (t-P) is the data of the P-th time point before the t time point, thereby constructing P explanatory variables { C s (t-p)} p＝1,2,...,P And a response variable C s (t)，
Figure FDA0003855166150000029
In all
Figure FDA00038551661500000210
Training samples;
s3.3.2) all explanatory variables
Figure FDA00038551661500000211
And all response variables
Figure FDA00038551661500000212
Respectively as input and output of the GRU prediction model of the gated cyclic unit neural network, and training to obtain the trained GRU prediction model and corresponding fitting value
Figure FDA00038551661500000213
Figure FDA00038551661500000214
Fitting values of the s-th subsequence at a time point t;
s3.3.3) rolling the data backwards by a time point, namely
Figure FDA00038551661500000215
As input to the trained GRU prediction model to obtain
Figure FDA00038551661500000216
Predicted value of time point
Figure FDA00038551661500000217
S3.3.4) assigning S +1 to S, judging whether S is more than S, if so, finishing the prediction of S subsequences, and thus obtaining the S subsequences from the moment P +1 to the moment S
Figure FDA00038551661500000218
Fitting value of time GRU prediction model
Figure FDA00038551661500000219
And in
Figure FDA00038551661500000220
Predicted value of time point
Figure FDA00038551661500000221
Otherwise, returning to the step S3.3.1;
s3.3.5) respectively calculating the wind power from the moment P +1 to the moment P +1 by using the formula (7) and the formula (8)
Figure FDA00038551661500000222
Integrated fitting value of time of day
Figure FDA00038551661500000223
And
Figure FDA00038551661500000224
initial predicted value of time point
Figure FDA00038551661500000225
Figure FDA00038551661500000226
Figure FDA0003855166150000031
S3.4) establishing a correction layer:
s3.4.1) extracting residual sequence of GRU prediction model according to formula (9)
Figure FDA0003855166150000032
Figure FDA0003855166150000033
In the formula (9), e (t) is the value of the residual sequence at the time t;
s3.4.2) utilizing residual error data { e (t-P) }of P time points before t time points of the residual error sequence p＝1,2,...,P Predicting data e (t) at the t moment, wherein e (t-p) is the data of the p-th moment point before the t moment point,
Figure FDA0003855166150000034
co-construction
Figure FDA0003855166150000035
Training samples;
s3.4.3) will
Figure FDA0003855166150000036
And
Figure FDA0003855166150000037
respectively serving as the input and the output of the SVR correction model and training to obtain a trained SVR correction model;
s3.4.4) rolling the data backwards by a time point, namely
Figure FDA0003855166150000038
As input of the trained SVR correction model, thereby obtaining
Figure FDA0003855166150000039
Residual prediction value of time point
Figure FDA00038551661500000310
S3.4.5) prediction layer Using equation (10)
Figure FDA00038551661500000311
Wind power initial prediction result of time point
Figure FDA00038551661500000312
Performing correction to obtain the final product
Figure FDA00038551661500000313
Final predicted value of time point
Figure FDA00038551661500000314
Figure FDA00038551661500000315
S3.5) establishing a feedback layer:
by wind farms
Figure FDA00038551661500000316
Actual measurement information of time point
Figure FDA00038551661500000317
Replacing the final predicted value at the corresponding time
Figure FDA00038551661500000318
And incorporating it into training samples to obtain an updated training sample set according to equations (11) and (12)
Figure FDA00038551661500000319
And updated total number of training samples
Figure FDA00038551661500000320
Figure FDA00038551661500000321
Figure FDA00038551661500000322
S3.6) will
Figure FDA00038551661500000323
Is assigned to
Figure FDA00038551661500000324
Will be provided with
Figure FDA00038551661500000325
Is assigned to
Figure FDA00038551661500000326
And then, repeating the step S3.2 to the step S3.5 until the prediction of the future M wind power values after the T moment is finished to obtain M wind power predicted values
Figure FDA00038551661500000327
S4) combining the classification criterion of the wind power climbing event in the step S2 and the predicted value of the wind power at the future M moment predicted by the multilayer cooperative prediction model in the step S3
Figure FDA0003855166150000041
And identifying the wind power climbing event, thereby realizing the real-time early warning of various wind power climbing events.
CN202010934286.6A 2020-09-08 2020-09-08 Multilayer collaborative real-time classification early warning method for wind power climbing event Active CN112070160B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010934286.6A CN112070160B (en) 2020-09-08 2020-09-08 Multilayer collaborative real-time classification early warning method for wind power climbing event

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010934286.6A CN112070160B (en) 2020-09-08 2020-09-08 Multilayer collaborative real-time classification early warning method for wind power climbing event

Publications (2)

Publication Number Publication Date
CN112070160A CN112070160A (en) 2020-12-11
CN112070160B true CN112070160B (en) 2022-11-08

Family

ID=73664324

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010934286.6A Active CN112070160B (en) 2020-09-08 2020-09-08 Multilayer collaborative real-time classification early warning method for wind power climbing event

Country Status (1)

Country Link
CN (1) CN112070160B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117408535B (en) * 2023-12-15 2024-04-12 国网山东省电力公司电力科学研究院 New energy climbing event rolling early warning method and system based on deep learning

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9098810B1 (en) * 2005-03-18 2015-08-04 Beyondcore, Inc. Recommending changes to variables of a data set to impact a desired outcome of the data set
CN106933778A (en) * 2017-01-22 2017-07-07 中国农业大学 A kind of wind power combination forecasting method based on climbing affair character identification
US11271398B2 (en) * 2018-07-13 2022-03-08 University Of Tennessee Research Foundation Voltage stability assessment, control and probabilistic power flow based on multi-dimensional holomorphic embedding techniques
CN110705768B (en) * 2019-09-26 2022-10-21 国家电网公司华北分部 Wind power generation power prediction optimization method and device for wind power plant

Also Published As

Publication number Publication date
CN112070160A (en) 2020-12-11

Similar Documents

Publication Publication Date Title
CN111562358B (en) Transformer oil gas content prediction method and system based on combined model
CN110221225B (en) Spacecraft lithium ion battery cycle life prediction method
CN107766990B (en) Method for predicting power generation power of photovoltaic power station
CN112633317A (en) CNN-LSTM fan fault prediction method and system based on attention mechanism
CN110837915B (en) Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning
CN112307676B (en) Wave height numerical prediction model result correction method
CN113344288B (en) Cascade hydropower station group water level prediction method and device and computer readable storage medium
CN116885691B (en) Wind power climbing event indirect prediction method
CN104036328A (en) Self-adaptive wind power prediction system and prediction method
CN113822418A (en) Wind power plant power prediction method, system, device and storage medium
CN114022311A (en) Comprehensive energy system data compensation method for generating countermeasure network based on time sequence condition
CN112070160B (en) Multilayer collaborative real-time classification early warning method for wind power climbing event
CN112651519A (en) Secondary equipment fault positioning method and system based on deep learning theory
CN110766215B (en) Wind power climbing event prediction method based on feature adaptive selection and WDNN
CN115908051A (en) Method for determining energy storage capacity of power system
CN110222910B (en) Active power distribution network situation prediction method and prediction system
CN111506868A (en) Ultrashort-term wind speed prediction method based on HHT weight optimization
CN111967660B (en) Ultra-short-term photovoltaic prediction residual error correction method based on SVR
CN113111592A (en) Short-term wind power prediction method based on EMD-LSTM
CN112948462A (en) Ultra-short-term wind speed prediction method based on improved singular spectrum analysis and Bp neural network
CN116167465A (en) Solar irradiance prediction method based on multivariate time series ensemble learning
CN115809732A (en) Distributed photovoltaic power generation power prediction method and system
CN115296298A (en) Wind power plant power prediction method
CN112070303B (en) Parameter-adaptive photovoltaic power ramp event hierarchical probabilistic prediction method
CN115965177A (en) Improved autoregressive error compensation wind power prediction method based on attention mechanism

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