CN105741192B - Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant - Google Patents

Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant Download PDF

Info

Publication number
CN105741192B
CN105741192B CN201610113700.0A CN201610113700A CN105741192B CN 105741192 B CN105741192 B CN 105741192B CN 201610113700 A CN201610113700 A CN 201610113700A CN 105741192 B CN105741192 B CN 105741192B
Authority
CN
China
Prior art keywords
wind
wind speed
dtw
pcc
sampling
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.)
Expired - Fee Related
Application number
CN201610113700.0A
Other languages
Chinese (zh)
Other versions
CN105741192A (en
Inventor
彭丽霞
杜杰
孙泓川
张琛
代刊
谌芸
毛冬艳
曹一家
陆金桂
刘玉宝
潘林林
刘月巍
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.)
Chongqing Yewen Data Technology Co ltd
Original Assignee
Nanjing University of Information Science and 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 Nanjing University of Information Science and Technology filed Critical Nanjing University of Information Science and Technology
Priority to CN201610113700.0A priority Critical patent/CN105741192B/en
Publication of CN105741192A publication Critical patent/CN105741192A/en
Application granted granted Critical
Publication of CN105741192B publication Critical patent/CN105741192B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • 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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Wind Motors (AREA)
  • Control Of Eletrric Generators (AREA)

Abstract

The invention discloses a short-term wind speed combined forecasting method for a wind turbine cabin of a wind power plant, which comprises the steps of respectively adopting a dynamic time warping method and a correlation coefficient method to carry out similarity analysis on wind speed subsequences which forecast the wind turbine cabin and all the wind turbine cabins and correspond to a certain time period day by day, extracting wind speed data of a plurality of subsequences with the most similar evolution, respectively establishing generalized regression neural network sub-model forecasting units based on the dynamic time warping method and the correlation coefficient method, carrying out global optimization on specific parameters of each sub-model by adopting a particle swarm algorithm, and taking the average value of forecasting results of the two sub-models as the final forecasting result of the combined forecasting method. According to the invention, the cabin wind speed of each wind turbine in the wind power plant is finely forecasted, so that the short-term output forecasting level of the whole wind power plant is effectively improved.

Description

Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant
Technical Field
The invention belongs to the technical field of wind motors of a wind power plant, and particularly relates to a short-term wind speed combined forecasting method for a cabin of a wind motor of the wind power plant.
Background
In order to effectively incorporate wind energy into a power grid, accurate prediction of the output of a wind power plant is extremely necessary and critical, wherein short-term prediction of 0 to 6 hours is of great significance for real-time scheduling of the power grid, ensuring of technical parameters related to power grid safety, such as power grid frequency, power and voltage balance, and the like.
The wind energy is a renewable clean energy, and has the advantages of flexible installation scale, high reliability of the wind power generator set, low manufacturing cost, simple operation and maintenance and the like. According to the monitoring situation of the wind power industry in 2014 published by the national energy agency of 2 months in 2015, the accumulated installed capacity of wind power in China reaches 9637 ten thousand kilowatts, accounts for 7 percent of the installed capacity of all power generation and accounts for 27 percent of the installed capacity of the wind power in the world by the end of 2014. In 2014, the wind power grid electricity quantity is 1534 hundred million kilowatts, which accounts for 2.78% of the total electricity generation quantity. The national energy agency of 12 months in 2014 publishes "strategic action plan for energy development (2014-2020), and the wind power installation is expected to reach 2 hundred million kilowatts in 2020. At present, wind power becomes the third main power supply in China after thermal power and hydroelectric power. With the continuous increase of installed capacity, the problem of electricity abandonment of wind power is always more prominent, and according to the statistics of the national energy agency, the average wind abandonment rate reaches 17% when the electricity abandonment amount of the whole country in 2012 is about 200 hundred million kilowatts; in 2013, the average wind abandon rate reaches 10% when the wind abandon amount of the whole country is about 150 hundred million kilowatts, and the latest statistics show that the average wind abandon rate is 7.5% when the wind abandon amount of the wind power reaches 86 hundred million kilowatts in 2014 to the end of 9 months. One important reason for causing the wind power electricity abandonment is that the wind intermittence causes the fluctuation and instability of the wind power to affect the wind power quality, and the electricity is abandoned in vain for ensuring the safety of a power grid. Based on this, the national energy agency publishes a temporary method for wind power plant power prediction and forecast management in 2011, and requires that all wind power plants which are connected to the grid and operate in China should establish a wind power prediction and forecast system and a power generation plan declaration working mechanism before 1 month and 1 day 2012 and start trial operation, and report a wind power prediction and forecast result according to requirements.
The conventional wind speed forecasting method of the wind power plant comprises a physical method and a statistical method, wherein the physical method is to obtain a timed, fixed-point and quantitative wind power forecasting output result of a numerical weather forecasting mode with high space-time resolution according to a refined numerical weather forecasting mode, and meanwhile, according to the actual operation condition of a wind power plant fan, various fan power generation influence factors are comprehensively considered, an output forecasting physical model is established, and the output forecasting of the wind power plant is carried out. The physical method does not need a large amount of measurement data, but requires accurate mathematical description of the atmospheric physical characteristics and the characteristics of the wind power plant, the equations are difficult to solve, the required data are large in amount, the calculation amount is large, the calculation time is long, and the difficulty and the cost for acquiring the data from a meteorological department are high, so that a statistical method is still commonly used in short-term wind power plant wind speed forecasting. At present, the statistical method mostly adopts a continuous method, a random time sequence method, a Kalman filtering method, a neural network method, a support vector machine and other methods according to historical data of a wind measuring tower of a wind power plant. The biggest disadvantage of forecasting only by relying on anemometer tower data is that a wind power plant is influenced by terrain, turbulence and the like, and the wind speed of an engine room at a wind turbine and the wind speed at a anemometer tower are obviously different, so that the large forecasting error is inevitably caused by only forecasting the output of the whole wind power plant by using the measured wind speed of the anemometer tower, and the method is irrelevant to a specific forecasting method. With the improvement of measurement technology and computer computing capability, the wind speed of a single generator cabin can be finely forecasted.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention aims to provide a short-term wind speed combined forecasting method for a wind turbine cabin of a wind power plant, and the wind speed of each wind turbine in the wind power plant is finely forecasted, so that the short-term output forecasting level of the whole wind power plant is effectively improved.
In order to achieve the technical purpose, the technical scheme of the invention is as follows:
a short-term wind speed combined forecasting method for a cabin of a wind turbine of a wind power plant is characterized by comprising the following steps:
(1) setting the number of the wind motors per day sampling wind speed data as M, and reading the original sampling wind speed set of all wind motor cabins in the wind power plant as { v (# i, j), wherein i is 1,2 …, M; j is 1,2, …, N, where v (# i, j) represents the nacelle wind speed value for wind turbine # i at sample point j, M is the total number of wind turbines, N is the total number of sample points,the total number of sampling days is
Figure GDA0000965246440000021
P is more than or equal to 1 and less than or equal to M of wind motor # p of the cabin wind speed to be forecasted, and the forecasting step length is LfAnd respectively adopting a dynamic time warping method and a Pearson correlation coefficient method to carry out sampling wind speed similarity measurement: the length of the similarity comparison is L, namely a wind motor # p is taken to backward push sampling wind speed data { v (# p, N-L +1), v (# p, N-L +2), …, v (# p, N) } of L length from a sampling point N, and the similarity comparison is respectively carried out according to a dynamic time warping method and a Pearson correlation coefficient method with the cabin sampling wind speed of a corresponding time period T every day in a period T before the D day of all wind motors in the wind farm, and the results of the two similarity comparison methods are respectively sorted from high to low according to the similarity measurement results;
(2) according to the results of the two similarity comparison methods obtained in the step (1), two GRNN forecast submodels are constructed, and three parameters of a training set P, a test set T and a smoothing coefficient S are required to be determined when the two GRNN forecast submodels are constructed;
constructing two simulation experiment sets to respectively determine the optimal parameters of the two GRNN forecast submodels, wherein the construction method of the simulation experiment sets comprises the following steps: the input of the training set P is a wind speed sequence set Q which is formed by the first I wind speed sequences most similar to { v (# P, N-L +1), v (# P, N-L +2), …, v (# P, N) } and aiming at the wind speed sequence length L for similarity comparison according to a dynamic time warping method or a Pearson correlation coefficient methodPThe output of the training set P is { v (# P, N-L +1), v (# P, N-L +2), …, v (# P, N) }; the test set T is the same as the training set P;
optimizing L, I and S by taking the self simulation errors of { v (# p, N-L +1), v (# p, N-L +2), …, v (# p, N) } as standards, wherein the optimal parameter of the GRNN sub-model established by aiming at the dynamic time warping method is LDTW、IDTWAnd SDTWThe optimal parameter of the GRNN sub-model established by aiming at the Pearson correlation coefficient method is LPCC、IPCCAnd SPCC；
(3) GRNN submodel established by dynamic time warping method according to LDTW、IDTWAnd SDTWBuilding a training set of GRNN
Figure GDA0000965246440000031
Test set
Figure GDA0000965246440000032
And a smoothing coefficient SDTWWherein, training set
Figure GDA0000965246440000033
Is the length L of the wind speed sequence for similarity comparison according to the dynamic time warping methodDTWObtained and { v (# p, N-L)DTW+1),v(#p,N-LDTW+2), …, v (# p, N) } most similar preceding IDTWWind speed sequence set formed by wind speed sequences
Figure GDA0000965246440000034
Training set
Figure GDA0000965246440000035
Output of { v (# p, N-L)DTW+1),v(#p,N-LDTW+2), …, v (# p, N) }; test set
Figure GDA0000965246440000036
Is inputted as
Figure GDA0000965246440000037
Each wind power generator is aimed at
Figure GDA0000965246440000038
L after each sampled wind speed sequencefWind speed sequence set and test set formed by sampling wind speed sequences
Figure GDA0000965246440000039
Is started from a sampling point N and has a step length L for a wind turbine # pfForecast wind speed sequence FDTWWherein
Figure GDA00009652464400000310
GRNN submodel established by Pearson correlation coefficient method according to LPCC、IPCCAnd SPCCBuilding a training set of GRNN
Figure GDA0000965246440000043
Test set
Figure GDA0000965246440000044
And a smoothing coefficient SPCCWherein, training set
Figure GDA0000965246440000045
Is the length L of the wind speed sequence for similarity comparison according to the Pearson correlation coefficient methodPCCObtained and { v (# p, N-L)PCC+1),v(#p,N-LPCC+2), …, v (# p, N) } most similar preceding IPCCWind speed sequence set formed by wind speed sequences
Figure GDA0000965246440000046
Training set
Figure GDA0000965246440000047
Output of { v (# p, N-L)PCC+1),v(#p,N-LPCC+2), …, v (# p, N) }; test set
Figure GDA0000965246440000048
Is inputted as
Figure GDA0000965246440000049
Each wind power generator is aimed at
Figure GDA00009652464400000410
L after each sampled wind speed sequencefWind speed sequence set and test set formed by sampling wind speed sequences
Figure GDA00009652464400000411
Is started from a sampling point N and has a step length L for a wind turbine # pfForecast wind speed sequence FPCCWherein
Figure GDA00009652464400000412
(4) Wind turbine # p starts from sampling point N with step length LfThe result of the forecast wind speed sequence is F ═ 0.5 (F)DTW+FPCC)。
Further, the definition of "a period T before day D" in step (1):
and if the sampling time interval of the wind speed sequence is less than one week, taking all data before the D-th day for similarity measurement, and otherwise, taking data in the same season one week or year by year before the D-th day for similarity measurement.
Further, in step (2), L, I and S are optimized by using a particle swarm optimization.
Further, a particle swarm algorithm is adopted to optimize L, I and S in a specific process:
(a) initializing population X ═ X1,X2,...,XW) Wherein W is the total number of particles and the ith particle is Xi＝(Li,Ii,Si) Particle velocity of Vi＝(v_Li,v_Ii,v_Si) Wherein L isi,Ii,SiA set of alternative solutions for parameters L, I and S;
(b) for each particle X in the populationiDetermined GRNN parameters, a simulation set was constructed from { v (# p, N-L)i+1),v(#p,N-Li+2), …, v (# p, N) } self simulation error calculates its fitness value, and takes the least fitness value as the optimization direction as the evaluation criterion to judge the quality of each particle, and records the particle XiThe extreme value of the current individual is P _ best (i), and the optimal individual in the group P _ best (i) is taken as the integral extreme value G _ best;
(c) each particle X in the populationiUpdating the position and the speed of the mobile terminal respectively;
Figure GDA0000965246440000041
Figure GDA0000965246440000042
in the formula: omega is the inertial weight, c1、c2Is an acceleration factor, and r1、r2Is distributed in [0,1 ]]The random number of (2);
(d) recalculating the objective function value of each particle at the moment, and updating P _ best (i) and G _ best;
(e) and (c) judging whether the maximum iteration number is reached, if so, ending the optimization process, otherwise, returning to the step (c).
Further, the inertia weight ω is 0.5, and the acceleration factor c1＝c2＝1.49445。
Adopt the beneficial effect that above-mentioned technical scheme brought:
(1) according to the method, on the basis of the characteristic that the daily change of the wind speed is closely related to the surface temperature of the underlying surface and 24-hour weak periodic change is presented, a dynamic time warping method and a Pearson correlation coefficient method are adopted to carry out similarity analysis on wind speed subsequences of the forecast wind motor cabins and all wind motor cabins in day-to-day corresponding time periods, wind speed data of a plurality of subsequences with the most similar evolution are extracted to construct a forecast model, the idea is that the method not only follows the weak periodic characteristic of the wind speed, but also similarity measurement is carried out in a wind motor historical sequence of a wind power place, so that the method is more beneficial to searching for the sequence which is most similar to the wind speed evolution of the wind motor to be forecasted. Simultaneously introducing two wind speed similarity comparison methods of a DTW (dynamic time warping) and Pearson correlation coefficient method, particularly a DTW method, which is essentially used for researching the similarity problem of two time sequences after stretching and shrinking, wherein the two wind speed time sequences for DTW (dynamic time warping) calculation can be unequal and accord with the actual working condition of wind speed measurement; meanwhile, the two wind speed time sequences are subjected to stretching/shrinking nonlinear mapping on a time axis, so that the method is more suitable for the actual working conditions of wind speeds which are similar in form but inconsistent in occurrence time and occurrence amplitude, and the theoretical basis of the method is particularly suitable for wind treatment. Experiments show that the system can effectively improve the forecasting precision of the short-term wind speed of the cabin of the wind turbine.
(2) The generalized regression neural network is used as a forecasting model, the generalized regression neural network has strong nonlinear mapping capability, a flexible network structure and high fault tolerance and robustness, when less sample data exists, the forecasting effect is good, and parameters needing to be adjusted only have a smooth coefficient S, so that the generalized regression neural network is particularly suitable for the application environment of the system.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a flow chart of a dynamic time warping method;
FIG. 3 is a schematic diagram illustrating GRNN submodel parameter optimization according to the present invention;
FIG. 4 is a schematic diagram of GRNN submodel forecasting in the present invention;
FIG. 5 is a schematic diagram of originally collected wind speed data of the nacelles of the wind turbines #1 to # 4;
FIG. 6 is a schematic diagram of Morlet wavelet spectrum analysis of originally acquired wind speed data of a wind turbine #1 cabin.
Detailed Description
The technical scheme of the invention is explained in detail in the following with the accompanying drawings.
The method of the present invention is illustrated in the flow chart of fig. 1, and includes the following steps.
Step one, data similarity judgment:
assuming that the number of the wind motor daily sampling wind speed data is M, reading in an original sampling wind speed set of all wind motor cabins in a wind power plant as { v (# i, j), wherein i is 1,2 …, M; j is 1,2, …, N, where v (# i, j) represents the nacelle wind speed value of wind turbine # i at sampling point j, M is the total number of wind turbines, N is the total number of sampling points, the total number of sampling days is
Figure GDA0000965246440000061
(
Figure GDA0000965246440000062
For a rounding down operation), N is also typically the predictor starting point. Wind motor # p (1 is less than or equal to) of cabin wind speed to be forecastedp is less than or equal to M), and the forecast step length is LfSampling wind speed similarity measurement is carried out by respectively adopting a Dynamic Time Warping (DTW) method and a Pearson Correlation Coefficient (PCC) method: the length of the similarity comparison is L, namely sampling wind speed data of L length is reversely pushed from a sampling point N by a wind motor # p, namely { v (# p, N-L +1), v (# p, N-L +2), …, v (# p, N) }, similarity comparison is carried out according to DTW and PCC methods with the cabin sampling wind speed of a corresponding time period T every day in a period T before the D day of all wind motors in the wind farm respectively, and the results are sorted according to the similarity measurement results from high to low.
Definition of "period T before day D" above: and if the sampling time interval of the wind speed sequence is less than one week, taking all data before the D-th day for similarity measurement, and otherwise, taking data in the same season one week or year by year before the D-th day for similarity measurement.
Wherein, 2 methods for comparing the wind speed evolution similarity are as follows:
1. dynamic time warping, as shown in fig. 2, assumes that two wind speed sequences X ═ X1,x2,…,xN}，Y＝{y1,y2,…,yNAnd N is the total number of the wind speed sequences. Initializing the sequence distance matrix dNNWherein d isNNEach element of (a) is:
Figure GDA0000965246440000071
in the matrix dNNIn (d), a set of adjacent matrix elements is referred to as a curved path, and is denoted as W ═ W1,w2,...,wKK-th element W of Wk＝(i,j)kThis path satisfies the following condition: (a) k is more than or equal to N and less than 2N-1; (b) w is a1＝(1,1)，wK(N, N); (c) for wk＝(i,j)，wk-1(i ', j') satisfies the conditions that i-i 'is not less than 0 and not more than 1, and j-j' is not less than 0 and not more than 1. On the basis of the above-mentioned technical scheme,
Figure GDA0000965246440000072
the DTW algorithm can be summarized as using the idea of dynamic programming to find a shortest path D from D (1,1) to D (N, N)The state transition equation is:
Figure GDA0000965246440000073
the smaller the DTW (X, Y), the higher the similarity of the sequence X, Y after stretching on the time axis.
2. Correlation coefficient method, two wind speed sequences X ═ X1,x2,…,xN}，Y＝{y1,y2,…,yNAnd N is the total number of sequences. The correlation coefficients of the sequences X, Y are:
Figure GDA0000965246440000074
wherein,
Figure GDA0000965246440000075
And
Figure GDA0000965246440000076
the average values of the sequences X and Y, respectively. The larger R (X, Y) is, the stronger the linear correlation of the sequences X and Y is.
Second step, GRNN submodel parameter optimization:
the method uses a Generalized Regression Neural Network (GRNN) to forecast the wind speed of the wind turbine cabin, and constructs two GRNN forecast submodels according to the result of similarity comparison of wind speed data in the step 1 by a DTW method and a PCC method respectively. Based on a GRNN algorithm, the GRNN is constructed by determining 3 parameters including a training set P, a test set T and a smoothing coefficient S, and the return value is a constructed neural network NET; here, two simulation experiment sets are constructed to determine the optimal parameters of the two sub-models, and the construction mode of the simulation experiment sets is as follows: the input of the training set P is the wind speed sequence length L for specific similarity comparison according to DTW method or PCC method, and the wind speed sequence set Q formed by the first I wind speed sequences most similar to { v (# P, N-L +1), v (# P, N-L +2), …, v (# P, N) } obtainedPThe output of the training set P is { v (# P, N-L +1), v (# P, N-L +2), …, v (# P, N) }; the test set T is the same as the training set P; by taking the simulation errors of { v (# p, N-L +1), v (# p, N-L +2), …, v (# p, N) } as standards, a Particle Swarm Optimization (PSO) is adopted for L, and,I and S are optimized, and the optimal parameter of the submodel established by aiming at the DTW method is LDTW、IDTWAnd SDTWThe optimal parameter of the sub-model established aiming at the PCC method is LPCC、IPCCAnd SPCC；
As shown in the result diagram of this step in fig. 3, for wind turbine # D, by constructing the optimization results of the simulation experiment set as the wind speed sequence of G1 sampling interval on the fourth sampling day of wind turbine # a, the wind speed sequence of G2 sampling interval on the second sampling day of wind turbine # B, and the wind speed sequence of G3 sampling interval on the first sampling day of wind turbine # C, G1, G2, and G3 have the smallest simulation error for H1;
the method comprises the following specific steps of optimizing L, I and S by adopting a particle swarm optimization:
(2-1) initializing population X ═ X1,X2,...,XW) Wherein W is the total number of particles and the ith particle is Xi＝(Li,Ii,Si) Particle velocity of Vi＝(v_Li,v_Ii,v_Si) Wherein L isi,Ii,SiA set of alternative solutions for parameters L, I and S;
(2-2) for each particle X in the populationiDetermined GRNN parameters, a simulation set was constructed from { v (# p, N-L)i+1),v(#p,N-Li+2), …, v (# p, N) } self simulation error calculates its fitness value, and takes the least fitness value as the optimization direction as the evaluation criterion to judge the quality of each particle, and records the particle XiThe extreme value of the current individual is P _ best (i), and the optimal individual in the group P _ best (i) is taken as the integral extreme value G _ best;
(2-3) Each particle X in the populationiUpdating the position and the speed of the mobile terminal respectively;
Figure GDA0000965246440000081
Figure GDA0000965246440000082
in the above formula, ω is an inertial weight, and may be 0.5; c. C1、c2As an acceleration factor, can take c1＝c21.49445; and r1、r2Is distributed in [0,1 ]]The random number of (2);
(2-4) recalculating the objective function value of each particle at the moment, and updating P _ best (i) and G _ best;
and (2-5) judging whether a convergence condition is met or not according to the maximum iteration times, if so, ending the optimization process, and otherwise, returning to the step (2-3).
Thirdly, forecasting GRNN submodels:
submodels built by DTW method, according to LDTW、IDTWAnd SDTWBuilding a training set of GRNN
Figure GDA0000965246440000091
Test set
Figure GDA0000965246440000092
And a smoothing coefficient SDTWWherein: training set
Figure GDA0000965246440000093
Is the length L of the wind speed sequence for similarity comparison according to the DTW methodDTWObtained and { v (# p, N-L)DTW+1),v(#p,N-LDTW+2), …, v (# p, N) } most similar preceding IDTWWind speed sequence set formed by wind speed sequences
Figure GDA0000965246440000094
Training set
Figure GDA0000965246440000095
Output of { v (# p, N-L)DTW+1),v(#p,N-LDTW+2), …, v (# p, N) }; test set
Figure GDA0000965246440000096
Is inputted as
Figure GDA0000965246440000097
Each wind power generator is aimed at
Figure GDA0000965246440000098
L after each sampled wind speed sequencefWind speed sequence set and test set formed by sampling wind speed sequences
Figure GDA0000965246440000099
Is started from a sampling point N and has a step length L for a wind turbine # pfForecast wind speed sequence FDTWWherein
Figure GDA00009652464400000910
Figure GDA00009652464400000911
Submodels built by PCC method, as LPCC、IPCCAnd SPCCBuilding a training set of GRNN
Figure GDA00009652464400000912
Test set
Figure GDA00009652464400000913
And a smoothing coefficient SPCCWherein: training set
Figure GDA00009652464400000914
Is the length L of the wind speed sequence for similarity comparison according to the PCC methodPCCObtained and { v (# p, N-L)PCC+1),v(#p,N-LPCC+2), …, v (# p, N) } most similar preceding IPCCWind speed sequence set formed by wind speed sequences
Figure GDA00009652464400000915
Training set
Figure GDA00009652464400000916
Output of { v (# p, N-L)PCC+1),v(#p,N-LPCC+2), …, v (# p, N) }; test set
Figure GDA00009652464400000917
Is inputted as
Figure GDA00009652464400000918
Each wind power generator is aimed at
Figure GDA00009652464400000919
L after each sampled wind speed sequencefWind speed sequence set and test set formed by sampling wind speed sequences
Figure GDA00009652464400000920
Is started from a sampling point N and has a step length L for a wind turbine # pfForecast wind speed sequence FPCCWherein
Figure GDA00009652464400000921
Figure GDA00009652464400000922
The results of this step are shown schematically in FIG. 4 for wind turbines # D, L after G1, G2 and G3fThe wind speed sequence set formed by the wind speed sequences G1 ', G2' and G3 'of the sampling wind speed sequences is used as the input of the test set, the output is the wind speed forecast value H1' of # D, and the forecast length is Lf。
Fourthly, combined forecasting:
wind turbine # p starts from sampling point N with a step length of LfThe result of the forecast wind speed sequence is F ═ 0.5 (F)DTW+FPCC)。
Specific test examples:
the total number of 274 wind motors in a certain wind power plant is shown in fig. 5, which is the original sampled wind speed data of 5 continuous days from #1 to #4 of the wind motors collected by the SCADA system in 2008, 10, 23 and 7, wherein each wind speed data is an average value in every 10 minutes, 144 wind speed data exist in each day, and the total number of the wind speed data is 720 in the figure. From the figure, it can be seen that the strong nonlinearity and randomness of the wind speed are shown, taking the wind turbine #1 as an example, the maximum value of the sampled wind speed data in the time period is about 18m/s, the minimum value is about 0.4m/s, the wind speed changes from about 13m/s to about 0.4m/s within 14 hours from the vicinity of the 90 th sampling point to the 170 th sampling point, and then the wind speed changes sharply to about 15 m/s. Besides, weak periodicity of wind speed with daily change can be seen, and Morlet wavelet spectrum analysis of the first 600 sampled wind speed data of the wind turbine #1 shows that the sequence contains quasi 144 sampling point (corresponding to 24 hours) periods, and only 24 hour periods in the sequence pass through a 95% red noise reliability check line, as shown in fig. 6, the weak periodicity of wind speed with daily change is also the starting point of the patent.
The application attaches 2 test examples, and example 1 shows a modeling process that a wind turbine #7 sets a new forecasting starting point every 6 sampling points from the 620 th sampling point, and simulates 11 forecasting starting points in a common mode, wherein the forecasting step length is 1; example 2 shows the results of comparison between 31 typhoon motors #7 to #37, and the method provided by the patent and the BP neural network extrapolation method, the GRNN neural network extrapolation method and the ARIMA time sequence method when the forecast step length is 1-6 for 11 forecast starting points simulated by experiment 1. The error evaluation criteria are mean square error MSE (mean Squared error) and mean Absolute error MAE (mean Absolute error),
Figure GDA0000965246440000101
wherein, YiAnd yiThe real value and the predicted value when the step length is i are respectively, and n is the total predicted step length.
Example 1
For the collected wind speed data of the #7 wind turbine, simulating a forecast starting point N ═ {620, 626, 632, 638, 644, 650, 656, 662, 668, 674, 680} and a forecast step length LfConstructing two simulation experiment sets to determine the optimal parameters of two sub-models, wherein the construction mode of the simulation experiment sets is as follows: according to the sub-model established by the DTW method or the PCC method, aiming at the wind speed sequence length L for similarity comparison, the wind speed sequence set Q formed by the first I wind speed sequences most similar to the { v (#7, N-L +1), v (#7, N-L +2), …, v (#7, N) } is obtainedPThe output of the training set P is { v (#7, N-L +1), v (#7, N-L +2), …, v (#7, N) }; test set T andthe training sets P are the same; let I e [2,6 ] by taking as a criterion the simulation error of { v (#7, N-L +1), v (#7, N-L +2), …, v (#7, N) } itself]、L∈[4,36]、S∈[0.1,0.5]L, I and S are optimized by adopting a particle swarm algorithm, the maximum iteration time is 30 times, a submodel established by the DTW method is named as DTW-PSO-GRNN, and the optimal parameter is LDTW、IDTWAnd SDTW(ii) a The sub-model established aiming at the PCC method is named as PCC-PSO-GRNN, and the optimal parameter is LPCC、IPCCAnd SPCCThe results are listed in table 1, the final combination model is named COM-PSO-GRNN, and the one-step prediction results and errors of the two sub-models established according to the respective optimal parameters and the corresponding combination model for each starting point are listed in table 2. It can be seen from table 1 and table 2 that the two sub-models optimized in table 1 have smaller wind speed sequence length for similarity comparison, and L isDTWHas a value of between 5 and 12, an average value of 7.7, LPCCThe value of (a) is between 5 and 20, the mean value is 9.5, which shows that the sequence length for similarity comparison is not too long for a highly nonlinear wind speed sequence, and simultaneously shows that the GRNN neural network still has good learning and generalization capability for shorter training set data; meanwhile, the smoothing coefficients of the GRNN of the two submodels in the table 1 are both close to 0.1, which shows that the smaller the smoothing coefficient of the GRNN is, the stronger the generalization capability of the GRNN is; the prediction results in Table 2 show that the error MSE of the DTW-PSO-GRNN is reduced by 7.33% compared with that of the PCC-PSO-GRNN, the precision of the combination model COM-PSO-GRNN is influenced by the PCC-PSO-GRNN and is between the two submodels, and the MSE is improved by 3.53% compared with that of the PCC-PSO-GRNN and is reduced by 3.67% compared with that of the DTW-PSO-GRNN.
TABLE 1
Figure GDA0000965246440000121
TABLE 2
Figure GDA0000965246440000122
Example 2
In order to verify the universality of the application, 31 wind turbines #7 to #37 are adopted, and when the forecast step length is 1-6 according to 11 forecast starting points simulated by experiment 1, the method provided by the patent comprises the comparison of two submodels and a combined model with a BP neural network extrapolation method, a GRNN neural network extrapolation method and an ARIMA time sequence method, and the results are listed in Table 3. As can be seen from Table 3, the prediction accuracy of COM-PSO-GRNN is the highest for both MSE and MAE error criteria, while the accuracy of the 3 extrapolation methods from high to low are GRNN neural network extrapolation, BP neural network extrapolation and ARIMA time series, respectively, which shows that:
1) short-term wind speed forecasting is feasible based on the similarity principle, and the effect is better than that of a method based on extrapolation;
2) the DTW-PSO-GRNN submodel is superior to the PCC-PSO-GRNN submodel because the dynamic time warping method essentially measures the nonlinear similarity of the wind speed time sequence on a time axis, and the correlation coefficient method is a linear correlation rule, and the wind speed is strong nonlinear;
3) the precision of the combined model is influenced by the precision of the sub-models, the precision of the combined model is not necessarily the highest, but the stationarity and the overall forecasting effect are always optimal;
4) the root cause of the larger error of the 3 extrapolation methods is the transmission of the error, and the error is increased sharply along with the increase of the prediction step length, wherein the prediction accuracy of the two neural network methods is far higher than that of an ARIMA time sequence method, which shows the superiority of a neural network model for a highly nonlinear wind speed sequence, and meanwhile, the prediction accuracy of the GRNN neural network extrapolation method is higher than that of a BP neural network extrapolation method, which shows that the generalization performance of the GRNN neural network is better than that of the BP neural network, so that the rationality of the method for adopting the GRNN neural network as the basis for wind speed prediction is also verified.
TABLE 3
Figure GDA0000965246440000131
The above embodiments are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modifications made on the basis of the technical scheme according to the technical idea of the present invention fall within the protection scope of the present invention.

Claims (5)

1. A short-term wind speed combined forecasting method for a cabin of a wind turbine of a wind power plant is characterized by comprising the following steps:
(1) setting the number of the wind motors per day sampling wind speed data as M, and reading the original sampling wind speed set of all wind motor cabins in the wind power plant as { v (# i, j), wherein i is 1,2 …, M; j is 1,2, …, N, where v (# i, j) represents the nacelle wind speed value of wind turbine # i at sampling point j, M is the total number of wind turbines, N is the total number of sampling points, the total number of sampling days is
Figure FDA0002873614240000014
P is more than or equal to 1 and less than or equal to M of wind motor # p of the cabin wind speed to be forecasted, and the forecasting step length is LfAnd respectively adopting a dynamic time warping method and a Pearson correlation coefficient method to carry out sampling wind speed similarity measurement: the length of the similarity comparison is L, namely a wind motor # p is taken to backward push sampling wind speed data { v (# p, N-L +1), v (# p, N-L +2), …, v (# p, N) } of L length from a sampling point N, and the similarity comparison is respectively carried out according to a dynamic time warping method and a Pearson correlation coefficient method with the cabin sampling wind speed of a corresponding time period T every day in a period T before the D day of all wind motors in the wind farm, and the results of the two similarity comparison methods are respectively sorted from high to low according to the similarity measurement results;
(2) according to the results of the two similarity comparison methods obtained in the step (1), two GRNN forecast submodels are constructed, and three parameters of a training set P, a test set T and a smoothing coefficient S are required to be determined when the two GRNN forecast submodels are constructed;
constructing two simulation experiment sets to respectively determine the optimal parameters of the two GRNN forecast submodels, wherein the construction method of the simulation experiment sets comprises the following steps: the input of the training set P is the first I wind speed sequences which are most similar to { v (# P, N-L +1), v (# P, N-L +2), …, v (# P, N) } and are obtained according to the wind speed sequence length L for similarity comparison by a dynamic time warping method or a Pearson correlation coefficient methodFormed wind speed sequence set QPThe output of the training set P is { v (# P, N-L +1), v (# P, N-L +2), …, v (# P, N) }; the test set T is the same as the training set P;
optimizing L, I and S by taking the self simulation errors of { v (# p, N-L +1), v (# p, N-L +2), …, v (# p, N) } as standards, wherein the optimal parameter of the GRNN sub-model established by aiming at the dynamic time warping method is LDTW、IDTWAnd SDTWThe optimal parameter of the GRNN sub-model established by aiming at the Pearson correlation coefficient method is LPCC、IPCCAnd SPCC；
(3) GRNN submodel established by dynamic time warping method according to LDTW、IDTWAnd SDTWBuilding a training set of GRNN
Figure FDA0002873614240000011
Test set
Figure FDA0002873614240000012
And a smoothing coefficient SDTWWherein, training set
Figure FDA0002873614240000013
Is the length L of the wind speed sequence for similarity comparison according to the dynamic time warping methodDTWObtained and { v (# p, N-L)DTW+1),v(#p,N-LDTW+2), …, v (# p, N) } most similar preceding IDTWWind speed sequence set formed by wind speed sequences
Figure FDA0002873614240000021
Training set
Figure FDA0002873614240000022
Output of { v (# p, N-L)DTW+1),v(#p,N-LDTW+2), …, v (# p, N) }; test set
Figure FDA0002873614240000023
Is inputted as
Figure FDA0002873614240000024
Each wind power generator is aimed at
Figure FDA0002873614240000025
L after each sampled wind speed sequencefWind speed sequence set and test set formed by sampling wind speed sequences
Figure FDA0002873614240000026
Is started from a sampling point N and has a step length L for a wind turbine # pfForecast wind speed sequence FDTWWherein
Figure FDA0002873614240000027
GRNN submodel established by Pearson correlation coefficient method according to LPCC、IPCCAnd SPCCBuilding a training set of GRNN
Figure FDA0002873614240000028
Test set
Figure FDA0002873614240000029
And a smoothing coefficient SPCCWherein, training set
Figure FDA00028736142400000210
Is the length L of the wind speed sequence for similarity comparison according to the Pearson correlation coefficient methodPCCObtained and { v (# p, N-L)PCC+1),v(#p,N-LPCC+2), …, v (# p, N) } most similar preceding IPCCWind speed sequence set formed by wind speed sequences
Figure FDA00028736142400000211
Training set
Figure FDA00028736142400000212
Output of { v (# p, N-L)PCC+1),v(#p,N-LPCC+2),…,v(#p,N)}；Test set
Figure FDA00028736142400000213
Is inputted as
Figure FDA00028736142400000214
Each wind power generator is aimed at
Figure FDA00028736142400000215
L after each sampled wind speed sequencefWind speed sequence set and test set formed by sampling wind speed sequences
Figure FDA00028736142400000216
Is started from a sampling point N and has a step length L for a wind turbine # pfForecast wind speed sequence FPCCWherein
Figure FDA00028736142400000217
(4) Wind turbine # p starts from sampling point N with step length LfThe result of the forecast wind speed sequence is F ═ 0.5 (F)DTW+FPCC)。
2. The method for combined forecasting of the short-term wind speed of the wind turbine engine room of the wind farm according to claim 1, characterized by comprising the following steps: the definition of "a period T before day D" in step (1) is,
and if the sampling time interval of the wind speed sequence is less than one week, taking all data before the D-th day for similarity measurement, and otherwise, taking data in the same season one week or year by year before the D-th day for similarity measurement.
3. The method for combined forecasting of the short-term wind speed of the wind turbine engine room of the wind farm according to claim 1, characterized by comprising the following steps: in step (2), L, I and S are optimized using a particle swarm algorithm.
4. The wind power plant wind motor cabin short-term wind speed combined forecasting method according to claim 3, characterized by comprising the following steps: the specific process of optimizing L, I and S by adopting a particle swarm algorithm in the step (2) comprises the following steps:
(a) initializing population X ═ X1,X2,...,XW) Wherein W is the total number of particles and the ith particle is Xi＝(Li,Ii,Si) Particle velocity of Vi＝(v_Li,v_Ii,v_Si) Wherein L isi,Ii,SiA set of alternative solutions for parameters L, I and S;
(b) for each particle X in the populationiDetermined GRNN parameters, a simulation set was constructed from { v (# p, N-L)i+1),v(#p,N-Li+2), …, v (# p, N) } self simulation error calculates its fitness value, and takes the least fitness value as the optimization direction as the evaluation criterion to judge the quality of each particle, and records the particle XiThe extreme value of the current individual is P _ best (i), and the optimal individual in the group P _ best (i) is taken as the integral extreme value G _ best;
(c) each particle X in the populationiUpdating the position and the speed of the mobile terminal respectively;
Figure FDA0002873614240000031
Figure FDA0002873614240000032
in the formula: omega is the inertial weight, c1、c2Is an acceleration factor, and r1、r2Is distributed in [0,1 ]]The random number of (2);
(d) recalculating the objective function value of each particle at the moment, and updating P _ best (i) and G _ best;
(e) and (c) judging whether the maximum iteration number is reached, if so, ending the optimization process, otherwise, returning to the step (c).
5. The wind power plant wind motor cabin short-term wind speed combined forecasting method according to claim 4The method is characterized in that: inertia weight ω 0.5, acceleration factor c1＝c2＝1.49445。
CN201610113700.0A 2016-02-29 2016-02-29 Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant Expired - Fee Related CN105741192B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610113700.0A CN105741192B (en) 2016-02-29 2016-02-29 Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610113700.0A CN105741192B (en) 2016-02-29 2016-02-29 Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant

Publications (2)

Publication Number Publication Date
CN105741192A CN105741192A (en) 2016-07-06
CN105741192B true CN105741192B (en) 2021-05-18

Family

ID=56248904

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610113700.0A Expired - Fee Related CN105741192B (en) 2016-02-29 2016-02-29 Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant

Country Status (1)

Country Link
CN (1) CN105741192B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108228428B (en) * 2018-02-05 2021-09-21 百度在线网络技术（北京）有限公司 Method and apparatus for outputting information
CN111537219B (en) * 2020-01-20 2021-11-02 内蒙古工业大学 Fan gearbox performance detection and health assessment method based on temperature parameters
CN112836359B (en) * 2021-01-21 2024-05-07 红塔烟草(集团)有限责任公司 Safety evaluation method for power supply and distribution equipment of cigarette production enterprises
CN115690335B (en) * 2023-01-04 2023-03-31 中集海洋工程有限公司 Safety testing method and system for offshore wind power emergency refuge cabin

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20140018497A (en) * 2012-08-01 2014-02-13 한국전력공사 Prediction method of short-term wind speed and wind power and power supply line voltage prediction method therefore
CN104331572A (en) * 2014-11-17 2015-02-04 南京工程学院 Wind power plant reliability modeling method considering correlation between air speed and fault of wind turbine generator
CN104463511A (en) * 2014-12-31 2015-03-25 哈尔滨工业大学 Wind speed intermittency quantitative depicting method based on turbine unit time starting-stopping frequency
CN104899665A (en) * 2015-06-19 2015-09-09 国网四川省电力公司经济技术研究院 Wind power short-term prediction method
CN105138729A (en) * 2015-07-24 2015-12-09 南京信息工程大学 Filling method based on PSO-GRNN (Particle Swarm Optimization-Generalized Regression Neural Network) for defect wind speed values of wind turbines in wind power plant

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20140018497A (en) * 2012-08-01 2014-02-13 한국전력공사 Prediction method of short-term wind speed and wind power and power supply line voltage prediction method therefore
CN104331572A (en) * 2014-11-17 2015-02-04 南京工程学院 Wind power plant reliability modeling method considering correlation between air speed and fault of wind turbine generator
CN104463511A (en) * 2014-12-31 2015-03-25 哈尔滨工业大学 Wind speed intermittency quantitative depicting method based on turbine unit time starting-stopping frequency
CN104899665A (en) * 2015-06-19 2015-09-09 国网四川省电力公司经济技术研究院 Wind power short-term prediction method
CN105138729A (en) * 2015-07-24 2015-12-09 南京信息工程大学 Filling method based on PSO-GRNN (Particle Swarm Optimization-Generalized Regression Neural Network) for defect wind speed values of wind turbines in wind power plant

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
风电场风机测量风速缺损值的组合填充模型;杜杰等;《电力自动化设备》;20150930;全文 *

Also Published As

Publication number Publication date
CN105741192A (en) 2016-07-06

Similar Documents

Publication Publication Date Title
CN110880789B (en) Economic dispatching method for wind power and photovoltaic combined power generation system
CN102419394B (en) Wind/solar power prediction method with variable prediction resolution
CN105741192B (en) Short-term wind speed combined forecasting method for wind turbine engine room of wind power plant
CN110334870B (en) Photovoltaic power station short-term power prediction method based on gated cyclic unit network
CN110009135B (en) Wind power prediction method based on width learning
Wu et al. Stratification-based wind power forecasting in a high-penetration wind power system using a hybrid model
CN105512766A (en) Wind power plant power predication method
CN112001537B (en) Short-term wind power prediction method based on gray model and support vector machine
CN105303250A (en) Wind power combination prediction method based on optimal weight coefficient
CN106203693A (en) A kind of system and method for Power Output for Wind Power Field climbing event prediction
CN115239029B (en) Wind power prediction method and system considering power time sequence and meteorological dependent characteristics
CN110866633A (en) Micro-grid ultra-short term load prediction method based on SVR support vector regression
CN106446440A (en) Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine
CN106611243A (en) Residual correction method for wind speed prediction based on GARCH (Generalized ARCH) model
Nana et al. Short-term wind speed prediction based on CNN_GRU model
CN113516271A (en) Wind power cluster power day-ahead prediction method based on space-time neural network
CN105825002A (en) Method for modeling dynamic equivalence of wind power farm based on dynamic grey-relevancy analysis method
Lodge et al. Short term wind speed prediction using artificial neural networks
CN105634018A (en) Random response surface method and interior point method based wind-power-plant-contained random optimal power flow solving method
CN103984986A (en) Method for correcting wind power ultra-short-period prediction of self-learning ARMA model in real time
CN112232714B (en) Deep learning-based risk assessment method for distribution network under incomplete structural parameters
CN111242371B (en) Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model
CN112734073A (en) Photovoltaic power generation short-term prediction method based on long and short-term memory network
Shi et al. Short-term photovoltaic power forecast based on long short-term memory network
Zhang et al. Multi-factor offshore short-term wind power prediction based on XGBoost

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Peng Lixia

Inventor after: Liu Yubao

Inventor after: Pan Linlin

Inventor after: Liu Yuewei

Inventor after: Du Jie

Inventor after: Sun Hongchuan

Inventor after: Zhang Chen

Inventor after: Acting Journal

Inventor after: Chen Yun

Inventor after: Mao Dongyan

Inventor after: Cao Yijia

Inventor after: Lu Jingui

Inventor before: Du Jie

Inventor before: Liu Yubao

Inventor before: Pan Linlin

Inventor before: Liu Yuewei

Inventor before: Peng Lixia

Inventor before: Sun Hongchuan

Inventor before: Zhang Chen

Inventor before: Acting Journal

Inventor before: Chen Yun

Inventor before: Mao Dongyan

Inventor before: Cao Yijia

Inventor before: Lu Jingui

GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20221226

Address after: No. 99, Middle Data Valley Road, Xiantao Data Street, Yubei District, Chongqing 401120

Patentee after: Chongqing Yewen Data Technology Co.,Ltd.

Address before: 210044, No. 219, Ning six road, Pukou District, Jiangsu, Nanjing

Patentee before: Nanjing University of Information Science and Technology

CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210518