CN109117992A - Ultra-short term wind power prediction method based on WD-LA-WRF model - Google Patents
Ultra-short term wind power prediction method based on WD-LA-WRF model Download PDFInfo
- Publication number
- CN109117992A CN109117992A CN201810839653.7A CN201810839653A CN109117992A CN 109117992 A CN109117992 A CN 109117992A CN 201810839653 A CN201810839653 A CN 201810839653A CN 109117992 A CN109117992 A CN 109117992A
- Authority
- CN
- China
- Prior art keywords
- model
- wind power
- lion
- data
- cub
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 241000282320 Panthera leo Species 0.000 claims abstract description 77
- 238000007637 random forest analysis Methods 0.000 claims abstract description 48
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 47
- 238000005457 optimization Methods 0.000 claims abstract description 14
- 238000003066 decision tree Methods 0.000 claims description 46
- 230000006870 function Effects 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 29
- 238000012360 testing method Methods 0.000 claims description 15
- 230000013011 mating Effects 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000009395 breeding Methods 0.000 claims description 7
- 241001269238 Data Species 0.000 claims description 6
- 230000002068 genetic effect Effects 0.000 claims description 6
- 230000003746 surface roughness Effects 0.000 claims description 5
- 230000001488 breeding effect Effects 0.000 claims description 4
- 230000035772 mutation Effects 0.000 claims description 4
- 238000011946 reduction process Methods 0.000 claims description 4
- 230000003862 health status Effects 0.000 claims description 3
- 238000013138 pruning Methods 0.000 claims description 3
- 230000008929 regeneration Effects 0.000 claims description 3
- 238000011069 regeneration method Methods 0.000 claims description 3
- 238000006467 substitution reaction Methods 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 241000894007 species Species 0.000 claims description 2
- 238000000354 decomposition reaction Methods 0.000 abstract description 22
- 230000005611 electricity Effects 0.000 abstract description 17
- 230000009467 reduction Effects 0.000 abstract description 6
- 230000006872 improvement Effects 0.000 abstract description 2
- 238000004458 analytical method Methods 0.000 description 10
- 238000005070 sampling Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 5
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 4
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000007619 statistical method Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000000053 physical method Methods 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 230000004931 aggregating effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000013277 forecasting method Methods 0.000 description 2
- 230000012010 growth Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007758 mating behavior Effects 0.000 description 2
- 239000011664 nicotinic acid Substances 0.000 description 2
- 238000013441 quality evaluation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000003749 cleanliness Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000000116 mitigating effect Effects 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000003472 neutralizing effect Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000007115 recruitment Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 230000027272 reproductive process Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000010415 tropism Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- 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)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Tourism & Hospitality (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Marketing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Genetics & Genomics (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Physiology (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Wind Motors (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of ultra-short term wind power prediction methods based on WD-LA-WRF model for belonging to ultra-short term wind power prediction technical field.The present invention considers the fluctuation and randomness of wind-power electricity generation, and wind series and wind power sequence are decomposed into using wavelet-decomposing method the subsequence of different frequency first, uses the parameter of lion algorithm optimization random forest, then to improve the precision of prediction model.MATLAB emulation finally is carried out using the model of building, each subsequence is predicted, prediction result is reconstructed, obtains final prediction result.The stationary signal of initial data is separated with non-stationary signal by decomposition to initial data and noise reduction, data is made to have more break-up value by WD-LA-WRF model proposed by the present invention.The present invention passes through the improvement of lion algorithm, optimizes the parameter of model, improves precision of prediction.The model and other single models compare, and have powerful generalization ability and robustness.It is predicted suitable for super short-period wind power.
Description
Technical field
The invention belongs to ultra-short term wind power prediction technical fields, and in particular to a kind of super based on WD-LA-WRF model
Short-term wind power prediction method.
Background technique
With the exhaustion of global fossil energy and the continuous growth of energy-consuming, it is each that Devoting Major Efforts To Developing new energy becomes the world
State copes with an important solution of " energy crisis ".Wind-power electricity generation as a kind of generation of electricity by new energy mode having a high potential,
Cleanliness without any pollution can effectively mitigation of climate change, raising energy security, promotion low-carbon economy increase, great commercialized development
Prospect, current each state are all increasing the research to wind-power electricity generation and its relevant technologies.But fluctuation and season due to wind
Difference round the clock, wind-power electricity generation has apparent intermittent and powerful fluctuation, after grid-connected, seriously affects power grid electricity
It can quality.Also, with increasing sharply for wind capacity integrated into grid, dispatching of power netwoks planning and dispatching of power netwoks operation are increasingly tired
Difficulty, wind-powered electricity generation are limited contradiction and more protrude, and the threat of Wind Power Generation on Power System is more serious.It is accurate to carry out in advance to wind power
Prediction, can alleviate the pressure of peak regulation electric system, effectively improve the ability of wind-electricity integration.Therefore wind power prediction is carried out
Have great importance.
It is divided according to research method, wind power forecasting method is broadly divided into physical method, statistical method and machine learning
Method；Because the main sample data of wind power output power is from physical factors such as wind speed, wind direction, temperature, humidity,
The basic thought of physical method is the weather information for obtaining numerical weather forecast and other physical messages as independent variable, structure
Build wind-power electricity generation power curve fitting real output.This method is predicted by matched curve, suitable for no history number
According to situation, such as newly-built wind power plant, but this method requires to collect more comprehensive data, and to data required precision
Height is not suitable for short-term forecast.
Statistical method establishes function model according to the changing rule of historical measurements mainly from statistical angle, right
The following wind power is predicted, such as auto regressive moving average (ARMA), autoregression quadrature rolling average (ARIMA), particle filter
Wave method, Markov chain, regression analysis, persistence model (PER) etc..Wherein, more the most commonly used is time series models and return
Return analysis, both models all have good prediction effect to stable data, but due to wind energy have it is apparent periodically and
Randomness shows strong non-stationary property.Compared with physical method, tool that statistical method changes without the concern for wind speed
Body process, modeling method is relatively easy, and precision of prediction is higher, but statistical method is needed by selecting suitable function model
Functional relation stringent between input and output is established, therefore, statistical model has significant limitation.
Machine learning algorithm is extracted the relationship between input and output, is then used with a large amount of historical data training pattern
Model predicts new sample data.Such algorithm mainly includes neural network, support vector machine, gray prediction, depth
Study etc..Such algorithm adaptability is stronger, model performance can be improved by constantly self-recision, so as to more preferably be located in
It manages the optimization problem of the complexity such as non-linear, non-differentiability but this method needs a large amount of data, analytical calculation process is increasingly complex,
And the disadvantages of there may be overfittings, poor robustness.
Wherein, neural network is a kind of more common algorithm, and it is excellent that there is stronger self-learning capability and high speed to seek for it
The ability of neutralizing mainly includes feedforward neural network and Feedback Neural Network.
With the development of intelligent predicting technology, support vector machines has received because of the advantages that its generalization ability is good, strong robustness
The favor of more scholars.At present using it is relatively broad be least square method support vector machines.
For wind-power electricity generation fluctuation and the stronger feature of randomness, researchers will use set empirical mode decomposition
(EEMD) former wind-power electricity generation sequence is resolved into multiple subsequences, then sub-sequences are predicted, and prediction result is reconstructed
For actual result, the accuracy predicted with raising.But this method is not still avoided that the shortcomings that modal overlap frequently occurs.
Therefore need one kind using MATLAB as tool, to decomposing former wind-power electricity generation sequence and carry out prediction model emulation
Model.
Summary of the invention
Aiming at the problem that mentioning in background technique, the invention discloses a kind of ultra-short term wind based on WD-LA-WRF model
Power forecasting method characterized by comprising
Step 1 is acquired and is pre-processed to data, acquires six sample datas of wind power output power, six sample datas
Are as follows: wind speed, wind power, wind direction, air pressure, atmospheric density, temperature and roughness of ground surface.Then sample data is normalized
Processing, nonumeric type data carry out category feature coding；
Step 2 carries out noise reduction process to pretreated sample data, then separately constitutes respectively with remaining sample data
From sample set enter step 3；
Step 3 carries out WD-LA-WRF model training using the Weighted random forest model of lion algorithm optimization；
Step 4 is obtained with step 3 using test sample as data source according to trained WD-LA-WRF model
WD-LA-WRF model is predicted respectively, and acquired results are reconstructed, and obtains final prediction result.
Data are first carried out nondimensionalization processing by normalized in the step 1, then initial data standardization is made it
It is between [0,1], standardization formula is as follows:
Wherein, xiIndicate a certain initial data, xmaxAnd xminRespectively represent the maximum value and minimum value in initial data；It will
Initial data can eliminate data dimension influence after standardization.
The step 2 is divided for following steps:
Step 21 wavelet mother function is the function or signal I (t), I (t) ∈ L for meeting following conditions2(R):
In formula:It is the Fourier transformation of I (t), t is the number of numerical value in sequence；
Wavelet mother function obtains continuous wavelet function by variation, the functional dependence in contraction-expansion factor w and translation parameters q,
Continuous wavelet function are as follows:
Wherein, w > 0, q ∈ R；
The continuous wavelet transform of definition signal O (t) are as follows:
In formula, O (t) ∈ L2(- ∞ ,+∞),For the conjugate function of I (t)；
To the wavelet transform of signal O (t) is defined as:
Wherein, j indicates an integer being gradually incremented by from 0 to infinity, to realize Fourier transformation in the form of power series
Variation；
After step 22 carries out wavelet transform to the signal O (t) of wind speed, with its in wind power output power sample data
Five sample sets of remaininging separately constitute respective sample set, remaining five sample set are as follows: wind direction, air pressure, atmospheric density, temperature and
Roughness of ground surface.
The step 3, which is divided, is:
Step 31 is set model parameter at random, is iterated using lion algorithm to all model parameters excellent
Change, obtains preferably model parameter；Model parameter are as follows: pruning threshold ε, decision tree number L, pretest sample number y and random spy
Levy variable number m；
Step 32 constructs Weighted random forest model
The input of Weighted random forest model is the sample data of wind power: wind speed, wind direction, temperature, relative humidity, gas
Pressure and surface roughness, export as wind power；
Step 33 is using the outer error Weighted random forest model of bag to progress estimation error
When having generated the performance of Weighted random forest model with data test outside bag, using data outside L group bag as input,
Generated Weighted random forest model is substituted into, L group predicted value is obtained, and the outer error OOBE of bag is calculated as follows:
Wherein, YlActual value is represented,Represent estimated value；
When error OOBE meets condition outside bag, terminate training, obtains random forest output as a result, being final prediction knot
Fruit.
The step 31, which is divided, is:
Before step 311, if objective function are as follows:
min g(x1,x2,...,xn) (14)
Wherein, n >=1；
Step 311: generating initial lion group
Lion group is initialized as 2n lions and is bisected into two groups, obtains a candidate population.Wherein, public lion knot
Structure isLioness structure isR is the length of solution vector；
Step 312: mating
Mating step introduces the interleaved mode based on double probability；In interleaved mode, the A of generationmAnd AfIt is produced by mating
Raw new cubU indicates cub number.Pass throughWithRandom selection two
A crosspoint, which carries out double probability intersections, can be generated four young baby groups
Mutation operation carries out random variation with Probability p, generates cubAfter the completion of intersecting and making a variation, cub species number
It is 8；
Existing 8 disaggregation are carried out gender grouping: male cub (A using K-means method by clusterm_cub) female cub
(Af_cub)；
By testing health status, according to the feasible solution that each individual represents, more one group of sum of slight of stature individual is killed,
That is the lesser individual of feasible solution；It is finally reached the purpose of Population Regeneration；
Step 313: territory defence
During territory is defendd, random initializtion travelling lion ψnomad, as new feasible solution.It is solved using new
Scheme αnomadAttack SambalionIf new solution is more preferable, it is compared with the solution of entire lion group, and will
Replace original lionNew lion will continue to mate, and original lion and young baby will be killed；Otherwise former lion
Son continue territory defence, cub grow up it is one-year-old, until cub grow up；
If g (x) is target function value, g (αpride) entire population value, calculation formula is as follows:
In formula, g (αmale) and g (αfemale) be respectively Sambalion and lioness value,WithIt is male respectively
Property cub and female cub value, | | αm_cub| | represent the number of male cub in population, and agematIt refers to mate
Age；
Step 314: territory adapter tube
The stage is taken in territory, finds the best solution in lioness and Sambalion respectively, substitutes solution inferior, into
Row mating, until reaching termination condition；It is substituted according to following standard:
Best Sambalion is selected according to the above standardWith best lioness
If η isBreeding number, ηstrenthFor the optimal breeding ability of lioness, 5 are typically set at, also, with lion group
Mating behavior gradually add one；If lioness is replaced, η is initialized as 0；If former lioness is replaced back, η is in former base
It adds up on plinth；
Step 315: whole process iteration is until reach termination condition；
Termination condition are as follows: GEN >=GENmax
Wherein GEN is the genetic algebra of lion group, GENmaxFor maximum genetic algebra；After reaching termination condition, select most
Excellent lion exports as preferably model parameter.
The step 32, which is divided, is:
Step 321 calculates sample set substitution Random Forest model
Step 322, the weight computing formula for carrying out pretest to every decision tree in Random Forest model are replaced are as follows:
Wherein, YlPretest sample wind power actual value is represented,Represent pretest sample wind power estimation value.
Step 323, the accuracy of every decision tree are weighted to obtain most as weight to every decision tree prediction result
Whole prediction result, calculation formula are as follows:
The invention has the benefit that
The present invention considers the fluctuation and randomness of wind-power electricity generation, uses wavelet-decomposing method by wind series and wind first
Power sequence is decomposed into the subsequence of different frequency, reinforces the day characteristic of wind series and wind power sequence.Then lion is used
The parameter of algorithm optimization random forest, to improve the precision of prediction model.Finally each subsequence is carried out using the model of building
Prediction, prediction result is reconstructed, final prediction result is obtained.
The combination WD-LA-WRF model realization proposed by the present invention mutual supplement with each other's advantages of each model, effectively reduces single
Model bring error.WD-LA-WRF model is by decomposition to initial data and noise reduction, by the stationary signal of initial data
It is separated with non-stationary signal, data is made to have more break-up value.
The present invention passes through the improvement of lion algorithm, optimizes the parameter of model, improves precision of prediction.The model and other
Single model comparison, has powerful generalization ability and robustness.It is predicted suitable for super short-period wind power.
Detailed description of the invention
Fig. 1 is wind power plant in a kind of ultra-short term wind power prediction embodiment of the method based on WD-LA-WRF model of the present invention
Wind series wavelet decomposition schematic diagram；
Fig. 2 is WD-LA-WRF model OOB error rate in the embodiment of the present invention；
Fig. 3 is WD-LA-WRF model actual value and predicted value comparison diagram in the embodiment of the present invention；
Fig. 4 is WD-LA-WRF model relative error in the embodiment of the present invention；
Fig. 5 is each model prediction result figure in the embodiment of the present invention；
Fig. 6 (a) is single decision tree training process in the embodiment of the present invention；
Fig. 6 (b) is random forest prediction model schematic diagram in the embodiment of the present invention；
Fig. 7 is lion algorithm interleaved mode in the embodiment of the present invention；
Fig. 8 is lion group defence figure in the embodiment of the present invention
Fig. 9 is WD-LA-WRFLA-WRF model schematic in the embodiment of the present invention；
Figure 10 is multi-scale wavelet decomposition schematic diagram in the embodiment of the present invention；
Specific embodiment
The embodiment of the present invention is described in conjunction with attached drawing；The embodiment of the present invention is based on wavelet decomposition and lion algorithm
Ultra-short term wind power prediction method --- the WD-LA-WRF model of weighted optimization random forest.
The present invention considers the fluctuation and randomness of wind-power electricity generation, uses wavelet-decomposing method by wind series and wind first
Power sequence is decomposed into the subsequence of different frequency, then uses the parameter of lion algorithm optimization random forest, to improve prediction
The precision of model.Finally each subsequence is predicted using the model of building, prediction result is reconstructed, is obtained final
Prediction result.In other words the present embodiment is broadly divided into three parts, is data prediction, noise reduction and prediction respectively, specifically:
(1) data prediction
The present invention has mainly carried out outlier processing and data normalized to data.Process of the initial data in generation
In will receive the influences of various extraneous factors, resulting error can also reduce the precision of prediction of wind power.Therefore, exist
Before usage history data are trained and test to model, for exceptional value, the value and subsequent time of last moment are used herein
The average value of value substitute the exceptional value.Unit and magnitude are different between the sample data of wind power, in order to avoid sample
Cause model convergence rate slow and training error mistake because the order of magnitude of each variable has big difference when notebook data is as input variable
Big situation needs that initial data is normalized.
(2) noise reduction part
Wavelet transformation be it is a kind of can time frequency analysis by a complicated signal decomposition at the signal in different frequency sections
Method.In view of wind-power electricity generation has certain periodicity (such as day characteristic), wind series and wind power sequence are generally non-
Stationary sequence carries out wavelet decomposition to wind series and wind power sequence, by by original series be decomposed into high fdrequency component and
Low frequency component, the sequence of different frequency can be made to show certain regularity, reinforce wind series and wind power sequence
Day characteristic.In the Time Series technology proposed at present, the noise removal capability of wavelet transformation compares empirical mode decomposition method
It is eager to excel.Therefore, the present invention selects wavelet-decomposing method to decompose original wind series and wind power sequence, by small
After wind series and wind power sequence are resolved into the signal of different frequency by wave conversion, wind power prediction can be reinforced
Accuracy.
(3) predicted portions
Lion algorithm is a kind of bionic Algorithm based on lion social action, is proposed by B.R.Rajakumar in 2012.
Lion group algorithm be a kind of parallel search method for not depending on particular problem, have adaptivity, collective search, it is heuristic with
The features such as machine search property.It is contemplated that existing for wind series and wind power sequence periodically, the present invention is with random forest
Basic forecast model, at the same it is weighted processing and lion algorithm optimization processing, using wavelet transformation to initial data into
Row noise reduction process, to reinforce the day characteristic of wind series and wind power sequence.
Simultaneously because the prediction of wind power is affected by many factors, such as wind speed, wind direction, climate condition, in order to
It can more accurately predict following 24 hours wind changed power situations, exception has been carried out to data in the embodiment of the present invention
Value processing and data normalized.
Specific step is as follows for the embodiment of the present invention:
The acquisition of step 1 data and pretreatment
Acquire wind power output power six sample datas, six sample datas are as follows: wind speed, wind power, wind direction, air pressure,
Atmospheric density, temperature and roughness of ground surface.Then sample data is normalized, nonumeric type data carry out classification
Feature coding；Step 1 in the present embodiment specifically:
Since initial data will receive the influence of various extraneous factors during generation, resulting error also can
Reduce the precision of prediction of wind power.Therefore, before usage history data are trained and test to model, for exceptional value,
The exceptional value is substituted using the average value of the value of last moment and the value of subsequent time herein.The sample data of wind power it
Between unit and magnitude it is different, in order to avoid when sample data is as input variable because the order of magnitude of each variable has big difference
And the situation that model convergence rate is slow and training error is excessive is caused, it needs that initial data is normalized.
Before being calculated, the relevant parameter rational change range according to specified in GB/T18710-2002 is to different first
Regular data carries out data cleansing.For data exception caused by extraneous factor, the average value of last moment and subsequent time are used
Substitute the exceptional value.
Then calculated, by data carry out nondimensionalization processing, by initial data standardization make its be in [0,1] it
Between, standardization formula is as follows:
Wherein, xiIndicate a certain initial data, xmaxAnd xminRespectively represent the maximum value and minimum value in initial data；It will
Initial data can eliminate data dimension influence after standardization.
Selection Inner Mongol wind power plant (A wind power plant) in the present embodiment on August 21, to 26 days 2017
Wind-power electricity generation data are that sample is analyzed, and sampling in every 5 minutes is primary, collect 288 points daily, form a sample set, altogether
There are 1728 sample points to carry out proof analysis.Wherein, 1440 data of first five day are arranged to training sample, last day
288 data be used as test sample.
Step 2 carries out noise reduction process to pretreated sample data, then separately constitutes respectively with remaining sample data
From sample set:
In the present embodiment, the sequence of wind speed and wind power is usually non-stationary series；Thus pass through small echo in step 2
After transformation is broken down into different frequency signals, the sequence of different frequency shows certain regularity.Wind series and wind
The wavelet decomposition of power sequence can make model proposed in this paper have more predictability.In the present embodiment first by taking wind speed as an example into
Row wavelet decomposition.
As shown in Figure 1, carrying out wavelet decomposition (noise reduction) to wind series, the details sequence and one of five different frequencies is obtained
A approximating sequence.By the original series of remaining five sample data directly as six subsequences, decomposed with original wind series
Six obtained subsequences correspond, and form six groups of data groups, and six groups of data for then obtaining processing are as LA-WRF mould
The input vector of type inputs six LA-WRF models arranged side by side.Six subsequences that wind Power Decomposition obtains are carrying out model instruction
When practicing, six LA-WRF models arranged side by side are also inputted simultaneously, but in model prediction, not in input model.WD-LA-WR model
Input vector be wind direction, temperature, relative humidity, air pressure, surface roughness and decompose wind series, output vector is wind speed
Sequence.Step 2 in the present embodiment specifically:
Wavelet transformation (WD) be it is a kind of can time-frequency by a complicated signal decomposition at the signal in different frequency sections
Analysis method.In view of wind-power electricity generation has certain periodicity (such as day characteristic), to wind series and wind power sequence into
Row wavelet decomposition reinforces wind series and wind power sequence by the way that original series are decomposed into high fdrequency component and low frequency component
Day characteristic.It is specific to be divided into following steps again:
Step 21 wavelet mother function is the function or signal I (t), I (t) ∈ L for meeting following conditions2(R):
In formula:It is the Fourier transformation of I (t), t is the number of numerical value in sequence.
Wavelet mother function obtains continuous wavelet function by variation, and the functional dependence is in contraction-expansion factor w (w > 0) and translation ginseng
Number q (q ∈ R), continuous wavelet function are as follows:
The continuous wavelet transform of definition signal O (t) are as follows:
In formula, O (t) ∈ L2(- ∞ ,+∞),For the conjugate function of I (t).
To the wavelet transform of signal O (t) is defined as:
Wherein, j indicates an integer being gradually incremented by from 0 to infinity, to realize Fourier transformation in the form of power series
Variation.
After step 22 carries out wavelet transform to the signal O (t) of wind speed, with its in wind power output power sample data
Five sample sets of remaininging separately constitute respective sample set, and 6 sample sets carry out step 3 respectively；Remaining five sample set are as follows: wind
To, air pressure, atmospheric density, temperature and roughness of ground surface.
In the present embodiment, data prediction and wavelet decomposition are carried out to original wind series and wind power sequence, it is first
First using southeastern coastal areas wind power plant on August 26th, 21 days 1 1728 data points of August in 2017 as signal sequence
Column input wavelet decomposition model, respectively obtains 5 high fdrequency components and a low frequency component, decomposition result is as shown in Figure 1.
Step 3 as shown in figure 9, using lion algorithm optimization Weighted random forest model carry out model training,
Breiman L introduced Bagging (Bootstrap aggregating) in 1994 in random forests algorithm
Algorithm is also named and haves no right double sampling.This method is to allow every decision tree to be all made of Bootstrap to repeat the methods of sampling from original
Training set is medium-scale to extract sub- training set (T1,T2,...,Tn), and for every decision tree construct an anticipation function (Γ (X,
T1),Γ(X,T2),...,Γ(X,Tn)).The final prediction result of random forest is exactly the set of every decision tree result；This
It uses Bootstrap algorithm to generate L training set sampling for L decision tree sampling in step, and is selected in each training set
Y pretest sample；L decision tree is generated respectively using the remaining sample of each training set；According to sample number and threshold epsilon
Compare, determines leaf node, and using the mode of its objective attribute target attribute as the classification results of the decision tree；Therefore step 3 specific steps
It is divided into again:
Step 31 is set model parameter at random, is iterated using lion algorithm to all model parameters excellent
Change, obtains preferably model parameter；Model parameter are as follows: pruning threshold ε, decision tree number L, pretest sample number y and random spy
Variable number m is levied, specifically:
Lion algorithm is a kind of bionic Algorithm based on lion social action, is proposed by B.R.Rajakumar in 2012.
Lion algorithm is evolved by the breeding of territory lion or territory lion competes (defence to the defence of flowing lion
) etc. compete modes realize the iteration and generation of optimal solution.Each lion represents a solution, passes through lion population
Variation continue to optimize solution, finally generate optimal solution, paper " The_Lions_Algorithm_A_New_Nature-
Inspired_Search " in provide pseudocode about lion algorithm, specifically:
The social action of lion algorithm simulation lion group, key step include generating initial population, mating variation, territory
Defence, territory adapter tube.The algorithm is finally obtained optimal solution, is finally obtained by constantly iteration and chess game optimization objective function
Optimal solution；Before step 311, if objective function are as follows:
min g(x1,x2,...,xn) (14)
Wherein, n >=1；
Step 311: generating initial lion group
In the initial stage of algorithm, lion group is initialized as 2n lions and is bisected into two groups, is obtained
To a candidate population.Wherein, public lion structure isLioness structure isR is the length of solution vector.
Step 312: mating
As shown in fig. 7, mating is an effectively method, it can lead to during iteration and search optimal solution
It crosses existing solution and generates new solution, mating operation is by intersecting, variation, and cluster kills sick, weak children
Son and etc. achieve the purpose that update lion group and maintain lion group it is stable.
Mating step introduces the interleaved mode (being intersected with two different probability) based on double probability, interleaved mode
As shown in Fig. 7.
The A of generationmAnd AfNew cub is generated by matingU indicates cub
Number.Pass throughWithRandomly choosing the double probability intersections of two crosspoints progress can be generated four young babies groups
Mutation operation carries out random variation with Probability p, generates cubAfter the completion of intersecting and making a variation, cub type
Number is 8.
Existing 8 disaggregation are carried out gender grouping: male cub (A using K-means method by clusterm_cub) female cub
(Af_cub)。
Finally, according to the feasible solution that each individual represents, killing sum more one by test health status (or target)
The slight of stature of group is individual (even if the lesser individual of feasible solution), is finally reached the purpose of Population Regeneration.And it is completed in population recruitment
Later, the age of cub is initialized as 0.
Step 313: territory defence
As shown in figure 8, will receive the attack of travelling lion in the reproductive process of lion group.At this point, Sambalion is in order to protect children
Son continues to occupy territory, defence can be unfolded to the attack of travelling lion, as shown in Figure 8.
During territory is defendd, random initializtion travelling lion ψnomad, as new feasible solution.It is solved using new
Scheme (αnomad) attack SambalionIf new solution is more preferable, it is compared with the solution of entire lion group,
And original lion will be replacedNew lion will continue to mate, and original lion and young baby will be killed；It is no
Then former lion continues territory defence, cub grow up it is one-year-old, until cub is grown up.
If g (x) is target function value, g (αpride) entire population value, calculation formula is as follows:
In formula, g (αmale) and g (αfemale) be respectively Sambalion and lioness value,WithIt is male respectively
Property cub and female cub value, | | αm_cub| | represent the number of male cub in population, and agematIt refers to mate
Age.
Step 314: territory adapter tube
The stage is taken in territory, finds the best solution in lioness and Sambalion respectively, substitutes solution inferior, into
Row mating, until reaching termination condition.It is substituted according to following standard:
Best Sambalion is selected according to the above standardWith best lioness
If η isBreeding number, ηstrenthFor the optimal breeding ability of lioness, 5 are typically set at, also, with lion group
Mating behavior gradually add one and (when initialization lion group, be set as 0).If lioness is replaced, η is initialized as 0.If former
Lioness is replaced back, and η adds up in original basis.
After completing above-mentioned steps, return step 312, until termination condition, termination condition is GEN >=GENmax；Wherein, GEN
For the genetic algebra of lion group, whole process iteration is until maximum genetic algebra GENmax, finally, after reaching termination condition,
The optimal lion selected exports as preferably model parameter.
Step 32 constructs Weighted random forest,
The present invention improves the Random Forest model in MATLAB, forms Weighted random forest model, and using should
Model is trained and predicts to sample data.Improved mode input is the sample data of wind power: wind speed, wind direction, temperature
Degree, relative humidity, air pressure and surface roughness export as wind power.This algorithm is as shown in Fig. 6 (a) and Fig. 6 (b).
Step 321 calculates sample set substitution Random Forest model
Random forest is a kind of integrated study sorting algorithm for having supervision, by the classifier group of more irrelevant decision trees
At.
The present embodiment using CART algorithm be as decision trees, specifically: CART is existed by Brimman L et al.
A kind of 1984 binary recursive subdivision algorithms proposed.Except for the leaf nodes, the algorithm is on each node by current sample
Collection is divided into two subsets.When dependent variable is a continuous variable, decision tree is a regression tree.What CART algorithm used
Segmentation rule influences the accuracy of decision tree；
Therefore the present invention is split node using variance reduction method；In node split, variance about subtraction use side
Poor formula come select it is optimal division and by minimum variance divide as criteria for classifying.In order to calculate the variance of each node, count
Calculate formula are as follows:
Wherein,Indicate average data, n indicates the quantity of data.
In Decision Tree Construction, according to minimum variance criteria, successive segmentation from top to bottom is needed, to generate classification
Rule.The classifying quality evaluation criterion of certain attribute is as follows:
arg max(Var-VarLeft-VarRight) (7)
Wherein: Var indicates variance, and VarLefts indicates left child node variance, and VarRight simultaneously indicates right child node variance.
When classifying quality evaluation meets condition, terminates training, form decision tree.Otherwise, initial step is returned to start to instruct
Practice.
By training, establish based on training set TnAnticipation function Γ (X, Tn).X indicates that input vector, Y indicate output
As a result.
Tn={ (X1,Y1),(X2,Y2),...,(Xn,Yn)},X∈R,Y∈R (8)
Breiman L introduced Bagging (Bootstrap aggregating) in 1994 in random forests algorithm
Algorithm is also named and haves no right double sampling.This method is to allow every decision tree to be all made of Bootstrap to repeat the methods of sampling from original
Training set is medium-scale to extract sub- training set (T1,T2,...,Tn), and for every decision tree construct an anticipation function (Γ (X,
T1),Γ(X,T2),...,Γ(X,Tn)).The final prediction result of random forest is exactly the set of every decision tree result.
A random training set is generated for every decision tree and constructs decision tree.In the generating process of decision tree, need
M attribute to be picked out at random from M all decision attributes as internal node and carries out branch, to complete decision tree
Growth.In the present invention, decision attribute is the sample data of wind power, including wind speed, wind direction, temperature, relative humidity, gas
Pressure and surface roughness, this m attribute are referred to as random character variable.Its production method generally uses Forestes-RI.
M=log2M+1 (9)
Wherein, m is random character variable number, and M is all decision attribute numbers.
Then, it forms random forest and executes algorithm.A large amount of decision tree is repeatedly generated using the above method, can be formed
Random forest.
The final prediction result of the model is sought using mean value method, is by the average value of every decision tree prediction result
As the final prediction result of random forest, calculation formula is as follows:
In formula, the as final prediction result of random forest.
Step 322, the weight computing formula for carrying out pretest to every decision tree in Random Forest model are replaced
In view of the predictive ability of every decision tree in random forest is all different, a part of decision tree predictive ability compared with
It is good, but another part decision tree predictive ability is poor, therefore the present invention is set according to the predictive ability of every decision tree
Its corresponding weight.
After the completion of the training of all decision trees, pretest is carried out to every decision tree, and calculate its weight using formula (12)；
This step specifically:
The training set of extraction is divided into conventional exercises sample and pretest sample, conventional exercises sample be used to decision tree into
Row training, pretest sample are used to test the decision tree that training terminates.It is as follows to calculate prediction accuracy formula:
Wherein, YlPretest sample wind power actual value is represented,Represent pretest sample wind power estimation value.
Step 323, the influence in order to reduce the poor decision tree of training effect in random forest use every decision tree
Accuracy is weighted to obtain final prediction result, calculation formula is as follows as weight to every decision tree prediction result:
When in the present embodiment to wind power prediction is carried out, by the WD wind speed subsequence decomposed and other samples
Data original sequence is the input variable of the model as the sample set that subsequence forms, and is input in decision tree；Wind power is
Result is exported for model.
The outer error of step 33 bag (OOBE) carries out estimation error
Weighted random forest model carrys out the generalization ability of assessment models using the outer error (OOBE) of bag.The outer data of bag refer to
When Bootstrap random sampling, the data do not chosen.In the performance for having generated Weighted random forest model with data test outside bag
When, using data outside L group bag as input, generated Weighted random forest model is substituted into, obtains L group predicted value, and as the following formula
Calculate the outer error OOBE of bag:
Wherein, YlActual value is represented,Represent estimated value.
When error meets condition outside bag, terminate training, obtains random forest output as a result, as final prediction result.
In the present embodiment, with 1440 of on August on August 25th, 21,1 2017 of southeastern coastal areas wind power plant
Data point carries out model training as sample, and the model obtained with training is on August 26th, 2,017 288 samples of the wind power plant
This point carries out wind power prediction.
The parameter of lion algorithm is initialized, setting parameter is as shown in table 1 below:
1 lion algorithm major parameter of table
| Parameter | Numerical value | Parameter | Numerical value |
| n | 5 | GENmax | 100 |
| agemat | 3 | Crossover probabilities | [0.2,0.6] |
| ηstrenth | 5 | Mutation probabilityp | 0.3 |
Then, optimization is iterated using parameter of the lion algorithm to Weighted random forest model.
It analyzes to obtain by OOBE, when OOBE reaches minimum value, the quantity of decision tree is optimal.Model in L=500,
It is preferable to restrain effect, and demonstrates WD-LA-WRF model convergence rate comparatively fast and globally optimal solution can be obtained, effect is such as
Shown in Fig. 2.
Therefore in the present embodiment, ε=0, L=500, y=500 are taken, when m=3, model has preferable prediction effect.
Step 4 wind power prediction
According to trained each model (each decision tree after obtaining preferably model parameter), come by data of test sample
Source is predicted respectively with WD-LA-WRF model, and acquired results is reconstructed, and final prediction result is obtained；This
In embodiment, trained model quantity is 6, in the present embodiment step 4 specifically:
1988, S.Mallat utilized the thought of multiresolution analysis, proposed Mallat algorithm, from spatially image
Illustrate the multi-resolution characteristics of small echo.Multi-scale wavelet decomposition is to be decomposed into signal by the change to scale factor
Approximate part and detail section, decomposable process are as shown in Figure 10.
Any signal can be 2 by resolution ratio-JLow frequency component and resolution ratio be 2-JHigh fdrequency component Perfect Reconstruction,
That is:
O (t)=D1+D2+D3+...+DJ+AJ (22)
Here wavelet transformation is carried out by taking wind series as an example.If original air speed data sequence is S, non-flat according to the algorithm
Steady air speed data sequence can be analyzed to the high-frequency signal d of different frequency1,d2,d3,...,dJWith a low frequency signal aJ, J is
Maximum decomposition level number.After the signal that decomposition obtains is reconstructed, details sequence D is obtained1,D2,D3,...,DJAnd approximating sequence
AJ, then have
S=D1+D2+D3+...+DJ+AJ (23)
S indicates reconstruction signal, AJThe approximate part (low frequency component) decomposed for J layers, DJThe detail portion decomposed for J layers
Divide (high fdrequency component).
In the present embodiment, the parameter obtained according to optimization, predicts the wind power of test set.WD-LA-WRF model pair
It is predicted by each subsequence that wavelet decomposition obtains, and prediction result is reconstructed, obtained on August 26th, 2017
288 data point wind power prediction results and relative error are as shown in Figure 3 and Figure 4；
From figs. 3 and 4 it can be seen that with WD-LA-WRF model Inner Mongolia regional August in 2017 sampled point on the 26th
Wind power is predicted that prediction curve fitting degree is preferable, and precision of prediction is higher.And the relative error of each future position
It is no more than 20%, it is relatively strong to illustrate that proposed model has within 10% for most future position relative errors
Generalization Capability and robustness.
After obtaining prediction curve, the present embodiment also chooses BP neural network (BP), supporting vector machine model herein
(SVM), four models such as Random Forest model (LA-RF) of Random Forest model (RF), lion algorithm optimization are to the wind-powered electricity generation
Field wind power is predicted, and the prediction result that prediction result is obtained with WD-LA-WRF model is compared and analyzed, as a result
As shown in Figure 5；
Can analyze to obtain from Fig. 5, WD-LA-WRF model proposed in this paper and BP neural network, SVM, RF, LA-RF,
It is compared Deng four models, for prediction result closer to true value, precision of prediction is higher.By to 288 test point prediction results and
The analysis of error can prove compared with other models, WD-LA-WRF model in terms of super short-period wind power prediction more
Stablize, robustness is more preferable.
And error analysis is carried out to the result of the present embodiment:
Error analysis is mainly that the deviation generated to required target is analyzed, it is for evaluation model precision of prediction
An important indicator.For the general applicability of verification algorithm, herein using average absolute percentage error (MAPE), it is square
Root error (RMSE), mean absolute error (MAE) and nonlinear function approximation goodness (R2) etc. four indexs carry out more each mould
The precision of prediction of type, to can more accurately evaluate the estimated performance of each model.The calculation formula of each index is as follows
It is shown:
Where,For predicted value, yiFor true value, n is the number of future position.During wind power prediction, MAPE,
RMSE, MAE variation have same tropism, and three error amounts are smaller, and prediction result is more accurate.Three each moulds of wind power plant
The index calculated result of type is as shown in table 2:
2. index calculation result table of table
As can be seen from Table 2: three error criterions (MAPE, RMSE, MAE) of proposed WD-LA-WRF model
It is the smallest in contrast model, and the goodness of fit is also relatively preferable, is 95.14%.In all models, BP nerve net
Network and SVM are two kinds of worst models of prediction effect.Error is made a concrete analysis of, the average absolute percentage of each model misses
Poor (MAPE) between 5%~29%, WD-LA-WRF model average absolute percentage error proposed by the invention is respectively
5.78%, it is lower than 10% and lower than other models.
LA-RF prediction effect is better than RF, SVM, BP model, but is slightly poorer than model WD-LA-WRF.Therefore, pass through four
The comprehensive analysis of index, available to draw a conclusion for short-term wind-electricity power precision of prediction: WD-LA-WRF > LA-RF > RF >
SVM>BP。
Claims (6)
1. a kind of ultra-short term wind power prediction method based on WD-LA-WRF model characterized by comprising
Step 1 is acquired and is pre-processed to data, acquires six sample datas of wind power output power, six sample datas are as follows: wind
Speed, wind power, wind direction, air pressure, atmospheric density, temperature and roughness of ground surface.Then sample data is normalized, it is non-
Numeric type data carries out category feature coding；
Step 2 carries out noise reduction process to pretreated sample data, then separately constitutes with remaining sample data respective
Sample set enters step 3；
Step 3 carries out WD-LA-WRF model training using the Weighted random forest model of lion algorithm optimization；
Step 4 is according to trained WD-LA-WRF model, using test sample as data source, with the obtained WD- of step 3
LA-WRF model is predicted respectively, and acquired results are reconstructed, and obtains final prediction result.
2. a kind of ultra-short term wind power prediction method based on WD-LA-WRF model according to claim 1, feature exist
In data are first carried out nondimensionalization processing by normalized in the step 1, then are in it initial data standardization
Between [0,1], standardization formula is as follows:
Wherein, xiIndicate a certain initial data, xmaxAnd xminRespectively represent the maximum value and minimum value in initial data；It will be original
Data can eliminate data dimension influence after standardization.
3. a kind of ultra-short term wind power prediction method based on WD-LA-WRF model according to claim 1, feature exist
In the step 2 is divided for following steps:
Step 21 wavelet mother function is the function or signal I (t), I (t) ∈ L for meeting following conditions2(R):
In formula:It is the Fourier transformation of I (t), t is the number of numerical value in sequence；
Wavelet mother function obtains continuous wavelet function by variation, and the functional dependence is in contraction-expansion factor w and translation parameters q, continuously
Wavelet function are as follows:
Wherein, w > 0, q ∈ R；
The continuous wavelet transform of definition signal O (t) are as follows:
In formula, O (t) ∈ L2(- ∞ ,+∞),For the conjugate function of I (t)；
To the wavelet transform of signal O (t) is defined as:
Wherein, j indicates an integer being gradually incremented by from 0 to infinity, to realize that Fourier transformation is changed in the form of power series；
After step 22 carries out wavelet transform to the signal O (t) of wind speed, in wind power output power sample data remaining five
A sample set separately constitutes respective sample set, remaining five sample set are as follows: wind direction, air pressure, atmospheric density, temperature and earth's surface are thick
Rugosity.
4. a kind of ultra-short term wind power prediction method based on WD-LA-WRF model according to claim 1, feature exist
In the step 3, which is divided, is:
Step 31 is set model parameter at random, is iterated optimization to all model parameters using lion algorithm, is obtained
To preferably model parameter；Model parameter are as follows: pruning threshold ε, decision tree number L, pretest sample number y and random character variable
Number m；
Step 32 constructs Weighted random forest model
The input of Weighted random forest model is the sample data of wind power: wind speed, wind direction, temperature, relative humidity, air pressure and ground
Surface roughness exports as wind power；
Step 33 is using the outer error Weighted random forest model of bag to progress estimation error
When having generated the performance of Weighted random forest model with data test outside bag, using data outside L group bag as input, substitute into
Generated Weighted random forest model obtains L group predicted value, and the outer error OOBE of bag is calculated as follows:
Wherein, YlActual value is represented,Represent estimated value；
When error OOBE meets condition outside bag, terminate training, obtains random forest output as a result, as final prediction result.
5. a kind of ultra-short term wind power prediction method based on WD-LA-WRF model according to claim 4, feature exist
In the step 31, which is divided, is:
Before step 311, if objective function are as follows:
min g(x1,x2,...,xn) (14)
Wherein, n >=1；
Step 311: generating initial lion group
Lion group is initialized as 2n lions and is bisected into two groups, obtains a candidate population.Wherein, public lion structure isLioness structure isR is solution
The length of vector；
Step 312: mating
Mating step introduces the interleaved mode based on double probability；In interleaved mode, the A of generationmAnd AfIt is generated by mating new
CubU indicates cub number.Pass throughWithRandomly choose two intersections
Point, which carries out double probability intersections, can be generated four young baby groups
Mutation operation carries out random variation with Probability p, generates cubAfter the completion of intersecting and making a variation, cub species number is 8
It is a；
Existing 8 disaggregation are carried out gender grouping: male cub (A using K-means method by clusterm_cub) female cub (Af _cub)；
By testing health status, according to the feasible solution that each individual represents, more one group of sum of slight of stature individual is killed
Row solves lesser individual；It is finally reached the purpose of Population Regeneration；
Step 313: territory defence
During territory is defendd, random initializtion travelling lion ψnomad, as new feasible solution.Use new solution
αnomadAttack SambalionIf new solution is more preferable, it is compared with the solution of entire lion group, and will replace former
The lion comeNew lion will continue to mate, and original lion and young baby will be killed；Otherwise former lion continue into
Row territory defence, cub grow up it is one-year-old, until cub grow up；
If g (x) is target function value, g (αpride) entire population value, calculation formula is as follows:
In formula, g (αmale) and g (αfemale) be respectively Sambalion and lioness value,WithIt is male children respectively
The value of son and female cub, | | αm_cub| | represent the number of male cub in population, and agematReferring to can mating age；
Step 314: territory adapter tube
The stage is taken in territory, finds the best solution in lioness and Sambalion respectively, solution inferior is substituted, is handed over
Match, until reaching termination condition；It is substituted according to following standard:
Best Sambalion is selected according to the above standardWith best lioness
If η isBreeding number, ηstrenthFor the optimal breeding ability of lioness, 5 are typically set at, also, with the friendship of lion group
Gradually add one with behavior；If lioness is replaced, η is initialized as 0；If former lioness is replaced back, η tires out in original basis
Add；
Step 315: whole process iteration is until reach termination condition；
Termination condition are as follows:
GEN≥GENmax
Wherein GEN is the genetic algebra of lion group, GENmaxFor maximum genetic algebra；After reaching termination condition, the optimal lion selected
Son is exported as preferably model parameter.
6. a kind of ultra-short term wind power prediction method based on WD-LA-WRF model according to claim 4, feature exist
In the step 32, which is divided, is:
Step 321 calculates sample set substitution Random Forest model；
Step 322, the weight computing formula for carrying out pretest to every decision tree in Random Forest model are replaced are as follows:
Wherein, YlPretest sample wind power actual value is represented,Represent pretest sample wind power estimation value；
Step 323, the accuracy of every decision tree are weighted to obtain final as weight to every decision tree prediction result
Prediction result, calculation formula are as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810839653.7A CN109117992B (en) | 2018-07-27 | 2018-07-27 | Ultra-short-term wind power prediction method based on WD-LA-WRF model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810839653.7A CN109117992B (en) | 2018-07-27 | 2018-07-27 | Ultra-short-term wind power prediction method based on WD-LA-WRF model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109117992A true CN109117992A (en) | 2019-01-01 |
| CN109117992B CN109117992B (en) | 2022-01-04 |
Family
ID=64862300
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810839653.7A Active CN109117992B (en) | 2018-07-27 | 2018-07-27 | Ultra-short-term wind power prediction method based on WD-LA-WRF model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109117992B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110532674A (en) * | 2019-08-27 | 2019-12-03 | 东北电力大学 | A kind of coal-fired power station boiler fire box temperature measurement method |
| CN110782071A (en) * | 2019-09-25 | 2020-02-11 | 天津大学 | Method for predicting wind power by convolutional neural network based on time-space characteristic fusion |
| CN112433507A (en) * | 2019-08-26 | 2021-03-02 | 电子科技大学 | LSO-LSSVM (least squares support vector machine) -based five-axis numerical control machine tool thermal error comprehensive modeling method |
| CN112734274A (en) * | 2021-01-20 | 2021-04-30 | 国家电网公司华中分部 | Mining and comprehensive evaluation method for low-carbon power grid operation leading influence factors |
| CN115167140A (en) * | 2022-07-27 | 2022-10-11 | 华北电力大学 | Multi-target random model prediction control strategy method and system for wind generating set |
| CN116562142A (en) * | 2023-05-11 | 2023-08-08 | 南京信息工程大学 | Meteorological element extension period forecasting method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268366A (en) * | 2013-03-06 | 2013-08-28 | 辽宁省电力有限公司电力科学研究院 | Combined wind power prediction method suitable for distributed wind power plant |
| US20150302313A1 (en) * | 2014-04-22 | 2015-10-22 | State Grid Corporation Of China | Method of predicating ultra-short-term wind power based on self-learning composite data source |
| CN107392363A (en) * | 2017-07-12 | 2017-11-24 | 河海大学 | A kind of CEEMD and random forest short-term wind power prediction method |
| CN107704953A (en) * | 2017-09-20 | 2018-02-16 | 河海大学 | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests |
-
2018
- 2018-07-27 CN CN201810839653.7A patent/CN109117992B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268366A (en) * | 2013-03-06 | 2013-08-28 | 辽宁省电力有限公司电力科学研究院 | Combined wind power prediction method suitable for distributed wind power plant |
| US20150302313A1 (en) * | 2014-04-22 | 2015-10-22 | State Grid Corporation Of China | Method of predicating ultra-short-term wind power based on self-learning composite data source |
| CN107392363A (en) * | 2017-07-12 | 2017-11-24 | 河海大学 | A kind of CEEMD and random forest short-term wind power prediction method |
| CN107704953A (en) * | 2017-09-20 | 2018-02-16 | 河海大学 | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112433507A (en) * | 2019-08-26 | 2021-03-02 | 电子科技大学 | LSO-LSSVM (least squares support vector machine) -based five-axis numerical control machine tool thermal error comprehensive modeling method |
| CN110532674A (en) * | 2019-08-27 | 2019-12-03 | 东北电力大学 | A kind of coal-fired power station boiler fire box temperature measurement method |
| CN110532674B (en) * | 2019-08-27 | 2022-08-12 | 东北电力大学 | Method for measuring temperature of boiler furnace of coal-fired power plant |
| CN110782071A (en) * | 2019-09-25 | 2020-02-11 | 天津大学 | Method for predicting wind power by convolutional neural network based on time-space characteristic fusion |
| CN112734274A (en) * | 2021-01-20 | 2021-04-30 | 国家电网公司华中分部 | Mining and comprehensive evaluation method for low-carbon power grid operation leading influence factors |
| CN112734274B (en) * | 2021-01-20 | 2023-11-03 | 国家电网公司华中分部 | Low-carbon power grid operation leading influence factor mining and comprehensive evaluation method |
| CN115167140A (en) * | 2022-07-27 | 2022-10-11 | 华北电力大学 | Multi-target random model prediction control strategy method and system for wind generating set |
| CN115167140B (en) * | 2022-07-27 | 2023-10-10 | 华北电力大学 | Multi-target random model predictive control strategy method and system for wind generating set |
| CN116562142A (en) * | 2023-05-11 | 2023-08-08 | 南京信息工程大学 | Meteorological element extension period forecasting method and device |
| CN116562142B (en) * | 2023-05-11 | 2023-12-01 | 南京信息工程大学 | Meteorological element extension period forecasting method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109117992B (en) | 2022-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109117992A (en) | Ultra-short term wind power prediction method based on WD-LA-WRF model | |
| Tian et al. | A novel wind speed forecasting system based on hybrid data preprocessing and multi-objective optimization | |
| Huang et al. | Two novel hybrid linear and nonlinear models for wind speed forecasting | |
| Zhao et al. | A self-organizing forecast of day-ahead wind speed: Selective ensemble strategy based on numerical weather predictions | |
| Tao et al. | Wind power prediction and pattern feature based on deep learning method | |
| CN109871977A (en) | Load forecasting method based on wavelet transform and the minimum support vector machines of optimization | |
| CN106022614A (en) | Data mining method of neural network based on nearest neighbor clustering | |
| Dou et al. | Hybrid model for renewable energy and loads prediction based on data mining and variational mode decomposition | |
| Zhang et al. | Wind speed prediction research considering wind speed ramp and residual distribution | |
| CN105608271A (en) | Decomposition and optimization based short-term wind speed time series prediction method | |
| CN101697167B (en) | Clustering-decision tree based selection method of fine corn seeds | |
| CN109583635A (en) | A kind of short-term load forecasting modeling method towards operational reliability | |
| Wang et al. | Ultra-short-term wind power forecasting based on deep belief network | |
| CN110796303B (en) | Short-term power load prediction method based on EWT and ODBSCAN | |
| CN115471008A (en) | Gas short-term load prediction method based on long-term and short-term memory neural network | |
| CN115965135A (en) | New energy prediction error modeling method and system based on naive Bayes classification | |
| Jiao et al. | Short-term building energy consumption prediction strategy based on modal decomposition and reconstruction algorithm | |
| CN110516792A (en) | Non-stable time series forecasting method based on wavelet decomposition and shallow-layer neural network | |
| CN113537556A (en) | Household short-term load prediction method based on state frequency memory network | |
| Chen et al. | Graph neural network-based wind farm cluster speed prediction | |
| Tao et al. | A hybrid wind power prediction method | |
| CN116822742A (en) | Power load prediction method based on dynamic decomposition-reconstruction integrated processing | |
| Wenting et al. | Short-term wind power forecasting model based on random forest algorithm and TCN | |
| Chen et al. | Short-term wind speed forecasting based on fuzzy C-means clustering and improved MEA-BP | |
| Zhang et al. | Wind power forecasting system with data enhancement and algorithm improvement |
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 |