CN105447510B - Fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM - Google Patents
Fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM Download PDFInfo
- Publication number
- CN105447510B CN105447510B CN201510764766.1A CN201510764766A CN105447510B CN 105447510 B CN105447510 B CN 105447510B CN 201510764766 A CN201510764766 A CN 201510764766A CN 105447510 B CN105447510 B CN 105447510B
- Authority
- CN
- China
- Prior art keywords
- lssvm
- wind speed
- formula
- fitness
- fluctuating wind
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Abstract
The present invention provides a kind of fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM, it with the simulation of ARMA Method for Numerical the following steps are included: generate the fluctuating wind speed time series sample of vertical space point first, and the fluctuating wind speed time series sample of spatial point is divided into training set, test set two parts, it is normalized respectively;Least square method supporting vector machine fluctuating wind speed prediction model is established, optimal LSSVM model parameter is found using artificial bee colony algorithm and is combined so that model predictive error is minimum.And it is compared using root-mean-square error, related coefficient and convergence number as evaluation index, and with the result of least square method supporting vector machine and least square method supporting vector machine (PSO-LSSVM) data driven technique of particle group optimizing.
Description
Technical field
The fluctuating wind speed prediction technique based on data-driven that the present invention relates to a kind of, it is specifically a kind of to be based on artificial bee
The fluctuating wind of group's optimization LSSVM (Least Squares Support Vector Machine, least square method supporting vector machine)
Fast prediction technique.
Background technique
It is former based on VC peacekeeping structural risk minimization to support vector machines (Support Vector Machine, SVM)
The machine learning algorithm based on data developed on the basis of reason has small sample, non-linear, high-dimensional, precision of prediction height
The characteristics of.SVM describes problem with a convex optimization problem when handling function approximation or forecasting problem, is based on
Mercer theorem, passes through Nonlinear MappingInput sample point is non-linearly reflected from the input space
It is mapped to high-dimensional feature space, then selects loss function, the minimum value of the loss function is solved in high-dimensional feature space,
The nonlinearity regression problem in sample space can be solved in feature space using the method for linear learning machine.
As that studies support vector machines gos deep into, that there is training algorithm speed is slow for support vector machines, algorithm is complicated and
The problems such as being difficult to realize, especially in Nonlinear Support Vector Machines, with drawing for the factors such as Lagrangian, kernel function
Enter, so that its calculating process is more complicated, causes to be that quadratic form optimisation technique can when solving dual problem the main reason for these
Can there is a problem of that training speed is slow.Because there is the calculating of a large amount of matrix in searching process, the big portion of algorithm is occupied
Between timesharing.
Least square method supporting vector machine (Least Squares Support Vector Machine, LSSVM) is
A kind of novel SVM that SuyKensJ.A.k is proposed, introduces the thought of least square, LSSVM in objective function using square
The insensitive loss function in tradition SVM is replaced with error loss function, replaces the inequality constraints in SVM with equality constraint,
So that standard SVM is solved quadratic programming problem and be converted into one group of linear relation of solution, the majorized function of LSSVM need to only solve linear etc.
Formula equation group, calculation amount are small, it is most important that the select permeability for avoiding penalty factor in SVM is asked to greatly simplify
Topic, improves learning rate.
The quality of LSSVM performance depends greatly on selected kernel functional parameter.RBF kernel function has high
Performance, therefore this invention selection is RBF kernel function, and the parameter of the LSSVM based on RBF kernel function is mainly concerned with core ginseng
Number σ and regularization parameter γ, the two parameter selections directly influence the performance of LSSVM.Traditional population (Particle
Swarm Optimization, PSO) algorithm, genetic algorithm (Genetic Algorithm, GA) and artificial neural network
The optimization algorithm of the intelligent extractions LSSVM parameter combinations (γ, σ) such as (Artifical Neural Network, ANN) algorithm is all deposited
The disadvantages of being easily trapped into slow local optimum, convergence rate, time-consuming, and artificial bee colony algorithm (Artificial Colony
Bee, ABC) it is the foraging behavior based on bee colony and the new optimization algorithm of one kind for proposing, there is fast convergence rate, precision
It is high, practical, reliable, the advantages that.In order to improve the performance of LSSVM, artificial bee colony algorithm is introduced, with based on artificial bee colony intelligence
Optimization algorithm carries out optimization of parameter choice to the kernel function of LSSVM, to improve the performance of LSSVM model.
Summary of the invention
The purpose of the present invention is to provide a kind of fluctuating wind speed prediction techniques based on artificial bee colony optimization LSSVM, solve
The problems such as precision of prediction of LSSVM is not high, and convergence rate is slow.Fluctuating wind speed time series sample is obtained by ARMA method numerical simulation,
As original systolic wind speed sample data.Sample data is divided into training set and test set, utilizes LSSVM data-driven skill
Art learns sample data, using the best parameter group (γ, σ) of ABC optimization algorithm intelligent extraction LSSVM, and establishes
Prediction model predicts other time sequence fluctuating wind speed.
Conceived according to foregoing invention, the present invention adopts the following technical solutions: a kind of arteries and veins based on artificial bee colony optimization LSSVM
Dynamic wind speed forecasting method, which is characterized in that itself the following steps are included:
Step 1: draw up fluctuating wind speed time series sample by ARMA method Numerical-Mode, as original systolic wind speed sample data,
The fluctuating wind speed time series sample of some spatial point is divided into training set, test set two parts, wind speed time series data is set
The Embedded dimensions m of prediction, is respectively normalized training set, test set;
Step 2: establishing LSSVM regressive prediction model, select RBF function as the kernel function of LSSVM, with Matlab
Software runs LSSVM model program, learns to training set;
Step 3: artificial bee colony algorithm is introduced into LSSVM regressive prediction model, LSSVM prediction model kernel function is set
The range σ ∈ [σ of parameter σ and regularization parameter γmin,σmax] and γ ∈ [γmin,γmax], nectar source number SN, largest loop time
Number MCN, threshold value Limit is abandoned;
Step 4: initial disaggregation x is randomly generatedij, using the root-mean-square error of prediction of wind speed and target wind speed as artificial bee
The fitness of group's algorithm, and calculate each initial solution xijFitness;
Step 5: bee is led to generate new explanation v in initial solution neighborhood searchij, according to greedy selection principle, even vijIt is suitable
Response is greater than xijFitness, then xij=vij, otherwise keep xijIt is constant;
Step 6: calculating all xijFitness, and calculate probability value Pi;
Step 7: following bee according to calculating probability value PiNectar source is selected, then carries out neighborhood search in selected nectar source
Generate new explanation vij, fitness is calculated, again according to greedy selection principle, if vijFitness be greater than xijFitness, then xij=
vij, otherwise keep xijIt is constant;
Step 8: judging whether there is the solution to be lost after Limit circulation, and if it exists, then generate one by search bee
A new explanation xijInstead of it;And store the best solution up to the present found;
Step 9: stopping iteration if current iteration number is greater than maximum times MCN;Otherwise the 4th step, Cycle are gone to
=Cycle+1;
Step 10: output best parameter group (γ, σ), and LSSVM prediction model is established, predict other time sequence
Air speed data.
Preferably, the formula of ARMA method is as follows in the first step:
In formula: U (t) is fluctuating wind speed;Ai、BjIt is the coefficient matrix of M × M rank AR and MA model respectively;X (t) is the rank of M × 1
Normal distribution white noise sequence;P is Autoregressive, q is that sliding returns skill;Correlation function is pungent by wiener-by power spectrum
Formula following formula calculates by the emperor himself:
By the matrixing to arma modeling formula, autoregressive coefficient A is solved respectivelyiWith sliding regression coefficient Bi, establish
Fluctuating wind speed expression formula;
Normalized processing formula are as follows:
In formula, xminIt is the minimum value of x, xmaxIt is the maximum value of x, it is using this formula that the range of x is whole to [0,1].
3, the fluctuating wind speed prediction technique according to claim 1 based on artificial bee colony optimization LSSVM, feature exist
In the kernel function that the second step LSSVM regressive prediction model selects is gaussian kernel function, the expression formula of RBF kernel function are as follows:
In formula, xi、xjFor training sample i-th, j element;γ is RBF kernel functional parameter.
Preferably, initial disaggregation x is randomly generated in the 4th stepijAnd the 8th generate in step new explanation formula it is as follows:
Wherein j ∈ { 1,2 ... D } is some component that D ties up feasible solution.
Preferably, the 5th step leads bee to generate new explanation v in initial solution neighborhood searchijFormula it is as follows:
Wherein j ∈ { 1,2 ... ... D }, k ∈ { 1,2 ... ... SN }, k be it is random generate and k ≠ i,Between [- 1,1]
Random number.
Preferably, the calculation formula of the 6th step fitness and probability value is as follows:
fitiIt is the fitness of i-th of feasible solution, the abundant degree of corresponding food source;fiCorresponding target function value.
Compared with prior art, the present invention has the following prominent advantages: traditional SVM preferably resolves previous
The problems such as small sample present in learning method, non-linear, overfitting, high dimension, local minimum, this knows that it in solution mode
And in Function Estimation problem splendid performance is not shown.But the SVM training complexity of standard is high, and to solve a band
The quadratic programming problem of constraint, and sample data is bigger, and solution quadratic programming problem is more complicated, and the trained time is longer.
LSSVM is developed on the basis of standard SVM, it replaces the insensitive loss function in SVM with quadratic loss function,
Inequality constraints is become into equality constraint, solution quadratic programming problem is avoided, improves training speed.Regularization parameter γ and
The core width cs of RBF function directly affect study and the generalization ability of LSSVM, therefore critically important to its parameter selection.Traditional pair
The method of LSSVM parameter extraction has grid data service, and cross-validation method is based on the Parameters selection of PSO (particle swarm algorithm), base
In the parameter selection method of GA (genetic algorithm), but these algorithms are time-consuming and laborious, and precision can not be guaranteed.Artificial bee colony
Algorithm (ABC) is introduced into as a kind of emerging optimization algorithm based on swarm intelligence, fast convergence rate and precision height
In LSSVM, optimal (γ, σ) parameter combination is extracted, to improve LSSVM performance, to improve the precision of prediction model.
Detailed description of the invention
Fig. 1 is ABC-LSSVM, PSO-LSSVM prediction of wind speed and actual wind speed comparison diagram at 30 meters;
Fig. 2 (a), Fig. 2 (b) are respectively the fitness comparison diagram of PSO-LSSVM, ABC-LSSVM forecasting wind speed at 30 meters;
Fig. 3 is ABC-LSSVM prediction of wind speed and actual wind speed correlation comparison diagram at 30 meters;
Fig. 4 is ABC-LSSVM, PSO-LSSVM prediction of wind speed and actual wind speed power spectrum comparison diagram at 30 meters;
Fig. 5 is ABC-LSSVM, PSO-LSSVM prediction of wind speed and actual wind speed comparison diagram at 50 meters;
Fig. 6 (a), Fig. 6 (b) are respectively the fitness comparison diagram of PSO-LSSVM, ABC-LSSVM forecasting wind speed at 50 meters;
Fig. 7 is ABC-LSSVM prediction of wind speed and actual wind speed correlation comparison diagram at 50 meters;
Fig. 8 is ABC-LSSVM, PSO-LSSVM prediction of wind speed and actual wind speed power spectrum comparison diagram at 50 meters;
Fig. 9 is ABC-LSSVM, PSO-LSSVM prediction of wind speed and actual wind speed comparison diagram at 60 meters;
Figure 10 (a), Figure 10 (b) are respectively the fitness comparison diagram of PSO-LSSVM, ABC-LSSVM forecasting wind speed at 60 meters;
Figure 11 is ABC-LSSVM prediction of wind speed and actual wind speed correlation comparison diagram at 60 meters;
Figure 12 is ABC-LSSVM, PSO-LSSVM prediction of wind speed and actual wind speed power spectrum comparison diagram at 60 meters;
Figure 13 is ABC-LSSVM forecasting wind speed flow chart.
Specific embodiment
Implementation of the invention is further described below in conjunction with attached drawing.
Artificial bee colony algorithm (Artificial Bee Colony, ABC) is a kind of meta-heuristic intelligent algorithm, it is opened
The foraging behavior in honeybee is sent out, for solving Numerical Optimization.ABC is mainly made of following three parts: leading bee
(Employed Bees), bee (Onlookers) and search bee (Scouts) are followed.In each cycle, lead bee quantity=with
With the quantity SN solved in bee quantity=group, the number of search bee is 1.
For ABC algorithm in solving optimization problem, the position of food source indicates a feasible solution of problem to be optimized, and honeybee is adopted
The process of honey namely searches the process of optimal solution.In algorithm, initialization generates SN solution, and each solution x firsti(i=1,
2 ... ... SN) it is all a D dimensional vector.When algorithm starts, leads bee first to carry out a field search to initial food source, compare
Search front and back nectar quality, if the food source nectar quality searched better than pervious, is replaced with new food source position
Otherwise old food source position keeps food source position constant.When it is all lead bee complete search after, existed by jive
Dancing area is communicated to food source nectar quality and follows bee.Follow bee according to lead bee provide information according to must probability select
It selects and bee is led to be followed, the probability that food source more abundant is selected is bigger.After following bee to choose and lead bee, also carry out primary
Neighborhood search, and retain preferable solution.Limit is known as abandoning threshold value being an important control parameter in ABC algorithm.It is false
Fixed food source nectar quality after given cycle-index Limit does not improve, and shows that this solution falls into part most
It is excellent, then solution is corresponding that bee is led to be changed into search bee with this.Assuming that the solution being abandoned is xi, then just being led to by search bee
It crosses formula (1) and a new solution is randomly generated to replace xi.ABC algorithm is exactly to eventually find optimal solution by duplicate search.
When initialization, the formula such as following formula (1) of feasible solution is randomly generated:
Wherein j ∈ { 1,2 ... D } is some component that D ties up feasible solution.
Honeybee records oneself optimal location up to the present, and neighborhood search is unfolded near current nectar source, generates one
It is such as following formula (2) that a new position, which replaces the formula of prior location:
Wherein j ∈ { 1,2 ... ... D }, k ∈ { 1,2 ... ... SN }, k be it is random generate and k ≠ i,Between [- 1,1]
Random number.
Adaptive value is calculated according to following formula in ABC algorithm:
According to fiWhether zero is greater than, such as following formula (3):
Bee is followed to select to lead the new probability formula of bee for such as following formula (4) in ABC algorithm:
fitiIt is the fitness of i-th of feasible solution, the abundant degree of corresponding food source;fiCorresponding target function value.
Wind series data are one group of data sequences changed over time, for given wind speed time series data { X
(t), t=1,2 ..., m }, least square method supporting vector machine need to carry out space to data sequence before selecting inputoutput data
Time series group, i.e., be converted into matrix form to find the relationship between data by reconstruct.For example, the wind speed X (t) of t moment can be by
The historical wind speed data X (t-1) at (t-1, t-2 ..., t-m) moment, X (t-2) ..., X (t-m) are predicted that m is insertion dimension
Number, prediction model can indicate are as follows: X (t)=f [X (t-1), X (t-2) ..., X (t-m)], Embedded dimensions m=10 in this patent.
It can establish the LSSVM wind speed regressive prediction model an of multiple input single output according to above-mentioned forecasting wind speed model.
The present invention is based on artificial bee colony optimization LSSVM fluctuating wind speed prediction technique the following steps are included:
Step 1: by ARMA, (Auto-Regressive and Moving Average Mode, autoregression sliding are flat
Equal model) method Numerical-Mode draws up fluctuating wind speed time series sample, as original systolic wind speed sample data, by some spatial point
Fluctuating wind speed time series sample is divided into training set, test set two parts, and the Embedded dimensions m that wind speed time series data is predicted is arranged,
Training set, test set are normalized respectively;For example, using 200 meters of Super Highs of ARMA numerical simulation along height side
To the fluctuating wind speed time series sample taken every the time step t=0.5s of 10000 steps of 10 meters of 20 spatial points, take wherein certain
1400 data of one spatial point, preceding 1000 data are training set, and rear 400 data are test set.The present embodiment difference
The Wind Velocity History sample curve of spatial point at 30 meters, 50 meters, 60 meters is taken to be trained, test, it is maximum number of iterations 200 times, embedding
Enter dimension m=10.
In the above-mentioned first step, the formula of ARMA method such as following formula (5) in the first step:
In formula: U (t) is fluctuating wind speed;Ai、BjIt is the coefficient matrix of M × M rank AR and MA model respectively;X (t) is the rank of M × 1
Normal distribution white noise sequence;P is Autoregressive, q is that sliding returns skill;Correlation function is pungent by wiener-by power spectrum
Formula following formula calculates by the emperor himself, such as following formula (6):
By the matrixing to arma modeling formula, autoregressive coefficient A is solved respectivelyiWith sliding regression coefficient Bi, establish
Fluctuating wind speed expression formula;
In the above-mentioned first step, formula, which is normalized, is, such as following formula (7):
Wherein, xminIt is the minimum value of input, xmaxThe maximum value of input, x are input value, and above formula is adjusted to the range of x
[0,1].
Step 2: establishing LSSVM regressive prediction model, select RBF function as the kernel function of LSSVM, with Matlab
Software runs LSSVM model program, learns to training set;
Wherein, the kernel function that second step LSSVM prediction model selects is gaussian kernel function (RBF), the expression of RBF kernel function
Formula is such as following formula (8):
In formula, xi、xjFor training sample i-th, j element;γ is RBF kernel functional parameter.
Step 3: artificial bee colony algorithm (ABC) is introduced into LSSVM regressive prediction model, LSSVM prediction model core is set
The range σ ∈ [σ of function parameter σ and regularization parameter γmin,σmax] and γ ∈ [γmin,γmax], nectar source number SN, maximum follow
Ring number MCN, threshold value Limit is abandoned;
Step 4: initial disaggregation x is randomly generated according to formula (1)ij, with the root-mean-square error of prediction of wind speed and target wind speed
(RMSE) fitness as ABC algorithm, and each solution x is calculated according to formula (3)ijFitness;
Initial disaggregation x is randomly generated in 4th stepijAnd the 8th generate new explanation in step formula such as following formula (9):
Wherein j ∈ { 1,2 ... D } is some component that D ties up feasible solution.
Step 5: bee is led to generate new explanation v in initial solution neighborhood search according to formula (2)ij, according to greedy selection principle, if
vijFitness be greater than xijFitness, then xij=vij, otherwise keep xijIt is constant;
5th step leads bee to generate new explanation v in initial solution neighborhood searchijFormula such as following formula (10):
Wherein j ∈ { 1,2 ... ... D }, k ∈ { 1,2 ... ... SN }, k be it is random generate and k ≠ i,Between [- 1,1]
Random number.
Step 6: calculating all xijFitness, according to formula (4) calculate probability value Pi;
The calculation formula such as formula (3) and formula (4) of the 6th step fitness and probability value.
Step 7: following bee according to PiNectar source is selected, new explanation v is generated according to formula (2) neighborhood searchij, fitness is calculated, according to
Greedy selection principle, if vijFitness be greater than xijFitness, then xij=vij, otherwise keep xijIt is constant;
Step 8: judging whether there is the solution to be lost, and if it exists, then search bee is according to formula after Limit circulation
(1) a new explanation x is generatedijInstead of it;Store the best solution up to the present found;
Step 9: stopping iteration if current iteration number is greater than maximum times MCN;Otherwise the 4th step, Cycle are gone to
=Cycle+1.
Step 10: combining (γ, σ) by the optimal parameter of output and establishing LSSVM prediction model, test set is carried out pre-
It surveys, predicts the air speed data of other time sequence, the fluctuating wind speed time series spectrum predicted;And use root-mean-square error
(RMSE), related coefficient (R) and convergence rate are compared analysis to LSSVM, PSO-LSSVM, ABC-LSSVM, as table 1 with
And table 2:
1 three kinds of method prediction result contrast tables of table
2 ABC-LSSVM, PSO-LSSVM convergence rate contrast table of table
The result that table 1 is shown can be seen that based on artificial bee colony (ABC) Optimized Least Square Support Vector (LSSVM)
The fluctuating wind speed value that predicts of prediction model and true value it is closer.Compared to LSSVM, PSO-LSSVM, ABC-LSSVM's
Mean square error minimum and maximum (R > 0.9, it is generally recognized that related coefficient is greater than 0.9, it is believed that has very strong correlation of related coefficient
Property).
The result that table 2 is shown can be seen that when the number of iterations is arranged at 200 times, most based on artificial bee colony (ABC) optimization
The small two forecasting wind speed model convergence rates for multiplying support vector machines (LSSVM) optimize least square branch with based on population (PSO)
The forecasting wind speed model convergence rate for holding vector machine (LSSVM), which is compared, will be higher by 8-10 times or more, and be based on artificial bee colony
(ABC) the forecasting wind speed model of Optimized Least Square Support Vector (LSSVM) is all restrained within 10 times mostly, is shown artificial
Bee colony optimization algorithm (ABC) has very strong robustness.
In above-mentioned second step, LSSVM regression forecasting algorithm description is as follows:
If giving l training sample: T={ (x1, y1) ... ... (xl, yl), wherein xi∈Rn, yi∈Rn, i=1,2 ..l,
Pass through Nonlinear MappingTraining input data is mapped in a high-dimensional feature space H, in spy
Optimal hyperlane is constructed in sign space H, sample data is fitted, and keeps the error of predicted value and target value minimum, LSSVM
The expression formula of regression fit problem is such as following formula (11):
Wherein, ω is weight vector, ξiIt is relaxation factor, b is deviation, and C is penalty factor, for carrying out to objective function
Control.
Lagrange function is introduced, defined function form such as following formula (12):
Wherein, αi, i=1,2,3 ... ... l are Lagrange multipliers.According to KKT condition, ω, b, ξ are carried out to L respectivelyi,
αiLocal derviation solves, and partial derivative is made to be equal to zero, obtains following equation and constraint condition, such as following formula (13):
LSSVM regressive prediction model is finally obtained, such as following formula (14)::
Invention introduces artificial bee colony optimization algorithm to the regressive prediction model of least square method supporting vector machine into parameter
Optimum choice improves precision of prediction and convergence rate.Therefore available conclusion: least square is optimized based on artificial bee colony
The fluctuating wind speed prediction (ABC-LSSVM) of support vector machines and the fluctuating wind based on particle group optimizing least square method supporting vector machine
Fast prediction technique (PSO-LSSVM) and LSSVM, which are compared, certain advantage.
Claims (6)
1. it is a kind of based on artificial bee colony optimization LSSVM fluctuating wind speed prediction technique, which is characterized in that itself the following steps are included:
Step 1: fluctuating wind speed time series sample is drawn up by autoregressive moving-average model ARMA method Numerical-Mode, as original arteries and veins
The fluctuating wind speed time series sample of some spatial point is divided into training set, test set two parts, wind is arranged by dynamic wind speed sample data
The Embedded dimensions m of fast time series data prediction, is respectively normalized training set, test set;
Step 2: establishing least square method supporting vector machine LSSVM regressive prediction model, core letter of the RBF function as LSSVM is selected
Number runs LSSVM model program with Matlab software, learns to training set;
Step 3: artificial bee colony algorithm is introduced into LSSVM regressive prediction model, LSSVM prediction model kernel functional parameter σ is set
With the range σ ∈ [σ of regularization parameter γmin,σmax] and γ ∈ [γmin,γmax], nectar source number, maximum cycle, abandon
Threshold value;
Step 4: initial disaggregation x is randomly generatedij, calculated using the root-mean-square error of prediction of wind speed and target wind speed as artificial bee colony
The fitness of method, and calculate each initial solution xijFitness;
Step 5: bee is led to generate new explanation v in initial solution neighborhood searchij, according to greedy selection principle, even vijFitness it is big
In xijFitness, then xij=vij, otherwise keep xijIt is constant;
Step 6: calculating all xijFitness, and calculate probability value Pi;
Step 7: following bee according to calculating probability value PiNectar source is selected, then neighborhood search is carried out in selected nectar source and generates newly
Solve vij, fitness is calculated, again according to greedy selection principle, if vijFitness be greater than xijFitness, then xij=xij, no
Then keep xijIt is constant;
Step 8: judging whether there is the solution to be lost, and if it exists, then produced by search bee after abandoning threshold value Limit times circulation
A raw new explanation xijInstead of it;And store the best solution up to the present found;
Step 9: stopping iteration if current iteration number is greater than maximum times MCN;Otherwise the 4th step, Cycle=are gone to
Cycle+1;
Step 10: output best parameter group, and LSSVM prediction model is established, predict the air speed data of other time sequence.
2. the fluctuating wind speed prediction technique according to claim 1 based on artificial bee colony optimization LSSVM, which is characterized in that
The formula of ARMA method is as follows in the first step:
In formula: U (t) is fluctuating wind speed;Ai、BjIt is the coefficient matrix of M × M rank AR and MA model respectively;X (t) is the rank normal state of M × 1
It is distributed white noise sequence;P is Autoregressive, q is that sliding returns skill;Correlation function passes through wiener-Xin Qingong by power spectrum
Formula following formula calculates:
By the matrixing to arma modeling formula, autoregressive coefficient A is solved respectivelyiWith sliding regression coefficient Bi, establish pulsation
Wind speed expression formula;
Normalized processing formula are as follows:
In formula, xminIt is the minimum value of x, xmaxIt is the maximum value of x, it is using this formula that the range of x is whole to [0,1].
3. the fluctuating wind speed prediction technique according to claim 1 based on artificial bee colony optimization LSSVM, which is characterized in that
The kernel function that the second step LSSVM regressive prediction model selects is gaussian kernel function, the expression formula of RBF kernel function are as follows:
In formula, xi, xjFor training sample i-th, j element;γ is RBF kernel functional parameter.
4. the fluctuating wind speed prediction technique according to claim 1 based on artificial bee colony optimization LSSVM, which is characterized in that
Initial disaggregation x is randomly generated in 4th stepijAnd the 8th generate in step new explanation formula it is as follows:
Wherein j ∈ { 1,2 ... D } is some component that D ties up feasible solution.
5. the fluctuating wind speed prediction technique according to claim 1 based on artificial bee colony optimization LSSVM, which is characterized in that
5th step leads bee to generate new explanation v in initial solution neighborhood searchijFormula it is as follows:
Wherein j ∈ { 1,2 ... D }, k={ 1,2 ... SN }, k be it is random generate and k ≠ i,Between [- 1,1]
Random number.
6. the fluctuating wind speed prediction technique according to claim 1 based on artificial bee colony optimization LSSVM, which is characterized in that
The calculation formula of the 6th step fitness and probability value is as follows:
fitiIt is the fitness of i-th of feasible solution, the abundant degree of corresponding food source;fiCorresponding target function value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510764766.1A CN105447510B (en) | 2015-11-11 | 2015-11-11 | Fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510764766.1A CN105447510B (en) | 2015-11-11 | 2015-11-11 | Fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105447510A CN105447510A (en) | 2016-03-30 |
CN105447510B true CN105447510B (en) | 2019-02-22 |
Family
ID=55557669
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510764766.1A Expired - Fee Related CN105447510B (en) | 2015-11-11 | 2015-11-11 | Fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105447510B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596118B (en) * | 2018-04-28 | 2021-04-23 | 北京师范大学 | Remote sensing image classification method and system based on artificial bee colony algorithm |
CN109271440B (en) * | 2018-08-20 | 2021-09-10 | 浙江大学 | Method for displaying large-scale multi-attribute stacked histogram in limited space |
CN110489790B (en) * | 2019-07-10 | 2022-09-13 | 合肥工业大学 | IGBT junction temperature prediction method based on improved ABC-SVR |
CN110619176A (en) * | 2019-09-18 | 2019-12-27 | 福州大学 | Aviation kerosene flash point prediction method based on DBN-RLSSVM |
CN111523635B (en) * | 2020-03-24 | 2023-07-28 | 南昌大学 | Harmonic detection method based on artificial bee colony algorithm combined with least square method |
CN112530529B (en) * | 2020-12-09 | 2024-01-26 | 合肥工业大学 | Gas concentration prediction method, system, equipment and storage medium thereof |
CN112883626A (en) * | 2021-01-15 | 2021-06-01 | 重庆邮电大学 | Electronic nose mixed gas concentration prediction method based on PSOABC-LSTM |
CN115907178B (en) * | 2022-11-30 | 2023-12-15 | 中国地质大学(武汉) | Clean ecosystem CO 2 Exchange amount prediction method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7557735B1 (en) * | 2006-09-05 | 2009-07-07 | Rockwell Collins, Inc. | Predictive and reactive windshear detection system and method |
CN104899431A (en) * | 2015-05-19 | 2015-09-09 | 上海大学 | LSSVM fluctuating wind speed prediction method based on integration of ant colony and particle swarm |
CN104933303A (en) * | 2015-06-10 | 2015-09-23 | 上海大学 | Method for predicting fluctuating wind speed based on optimization-based multiple kernel LSSVM (Least Square Support Vector Machine) |
CN104951798A (en) * | 2015-06-10 | 2015-09-30 | 上海大学 | Method for predicting non-stationary fluctuating wind speeds by aid of LSSVM (least square support vector machine) on basis of EMD (empirical mode decomposition) |
-
2015
- 2015-11-11 CN CN201510764766.1A patent/CN105447510B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7557735B1 (en) * | 2006-09-05 | 2009-07-07 | Rockwell Collins, Inc. | Predictive and reactive windshear detection system and method |
CN104899431A (en) * | 2015-05-19 | 2015-09-09 | 上海大学 | LSSVM fluctuating wind speed prediction method based on integration of ant colony and particle swarm |
CN104933303A (en) * | 2015-06-10 | 2015-09-23 | 上海大学 | Method for predicting fluctuating wind speed based on optimization-based multiple kernel LSSVM (Least Square Support Vector Machine) |
CN104951798A (en) * | 2015-06-10 | 2015-09-30 | 上海大学 | Method for predicting non-stationary fluctuating wind speeds by aid of LSSVM (least square support vector machine) on basis of EMD (empirical mode decomposition) |
Non-Patent Citations (1)
Title |
---|
基于人工蜂群算法优化LSSVM的蒸汽干度软测量;杨日光等;《化工机械》;20130415;第226页-第229页 |
Also Published As
Publication number | Publication date |
---|---|
CN105447510A (en) | 2016-03-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105447510B (en) | Fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM | |
Wu et al. | Evolving RBF neural networks for rainfall prediction using hybrid particle swarm optimization and genetic algorithm | |
Uzlu et al. | Estimates of hydroelectric generation using neural networks with the artificial bee colony algorithm for Turkey | |
CN107665230A (en) | Training method and device for the users' behavior model of Intelligent housing | |
CN109142946A (en) | Transformer fault detection method based on ant group algorithm optimization random forest | |
CN101414366B (en) | Method for forecasting electric power system short-term load based on method for improving uttermost learning machine | |
CN109919316A (en) | The method, apparatus and equipment and storage medium of acquisition network representation study vector | |
CN102411687B (en) | Deep learning detection method of unknown malicious codes | |
CN109101938A (en) | A kind of multi-tag age estimation method based on convolutional neural networks | |
CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
CN113762603B (en) | Photovoltaic base station short-term photovoltaic power prediction method based on improved sparrow algorithm optimization | |
CN105654136B (en) | A kind of extensive remote sensing image Motion parameters method based on deep learning | |
CN109614495A (en) | A kind of associated companies method for digging of combination knowledge mapping and text information | |
CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
CN112819192A (en) | Method for predicting short-term power load of RF _ GRU network based on swarm algorithm optimization | |
CN106779219A (en) | A kind of electricity demand forecasting method and system | |
CN110516537A (en) | A kind of face age estimation method based on from step study | |
CN111008790A (en) | Hydropower station group power generation electric scheduling rule extraction method | |
CN110472840A (en) | A kind of agricultural water conservancy dispatching method and system based on nerual network technique | |
Wang et al. | A new approach of obtaining reservoir operation rules: Artificial immune recognition system | |
CN107807919A (en) | A kind of method for carrying out microblog emotional classification prediction using random walk network is circulated | |
CN106845696B (en) | Intelligent optimization water resource configuration method | |
CN116012094A (en) | Depth map matching network and fashion collocation recommendation method based on same | |
CN110502964A (en) | A kind of pedestrian's recognition methods again of unsupervised data-driven | |
CN111062511B (en) | Aquaculture disease prediction method and system based on decision tree and neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190222 Termination date: 20211111 |
|
CF01 | Termination of patent right due to non-payment of annual fee |