CN108879656A - A kind of Short-Term Load Forecasting Method integrated based on sub-sample SVR - Google Patents
A kind of Short-Term Load Forecasting Method integrated based on sub-sample SVR Download PDFInfo
- Publication number
- CN108879656A CN108879656A CN201810590419.5A CN201810590419A CN108879656A CN 108879656 A CN108879656 A CN 108879656A CN 201810590419 A CN201810590419 A CN 201810590419A CN 108879656 A CN108879656 A CN 108879656A
- Authority
- CN
- China
- Prior art keywords
- sample
- svr
- sub
- integrated
- short
- 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
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
Abstract
The invention belongs to the quick analysis technical fields of computer data, disclose a kind of Short-Term Load Forecasting Method integrated based on sub-sample SVR；The method integrated using sub-sample support vector regression, sub-sampling strategy realize the collateral learning of support vector machines, ensure that each single SVR has enough diversity, reduce the loss amount of information；The group's optimized learning algorithm for selecting the one kind integrated based on single SVR new, guaranteeing that each SVR is integrated, there is enough intensity to predict short term data.There is the statistical inference property of classical U- statistic the present invention is based on the integrated approach of small size sub-sample, the introducing of small-scale sub-sample strategy effectively reduces the complexity of estimator；Computational accuracy and efficiency are improved simultaneously, and returns to preferable confidence interval simultaneously.SSVRE mode of the invention is easy to be transplanted in parallel computation frame.
Description
Technical field
The invention belongs to the quick analysis technical field of computer data more particularly to it is a kind of based on sub-sample SVR integrate
Short-Term Load Forecasting Method (SSVRE).
Background technique
Short-term electric load prediction (STLF) plays a crucial role in operation power, and short-term load forecasting is (such as
Half an hour electric load) problem will generate a large amount of data in real time；Therefore, data can be quickly handled, reasonable point estimation is provided
Prediction and a great problem that Estimating Confidence Interval is that Utilities Electric Co. faces.Currently, the prior art commonly used in the trade is such：It is existing
There is one Che J, Wang J, Tang Y.Optimal training subset in a support vector of technology
Regression electric load forecasting model.Applied Soft Computing 2012,12 (5):
1523-1531. is based on VC dimension theory and " core " technology, is the multiple linear in high-dimensional feature space by SVR model indirect reformer
Regression problem, wherein being optimized by solving quadratic programming problem to the solution.However, the computation complexity of this problem is O
(N3) (number that N is training data point).Therefore, the opening that complexity has become statistical machine learning field how is reduced
Property problem.Two Che J, Yang Y L, Li L, et al.A modified support vector of the prior art
regression:Integrated selection of training subset and model.Applied Soft
Computing 2017,53:308-322. selects being associated between training subset selection in view of model, and attempting will training
The selection course of collection and model combines, and proposes nested particle swarm optimization algorithm (NPSO), then adaptively and periodical
Estimate the region of search of the optimized parameter of SVR in ground；Therefore, the complicated SVR for being related to large scale training data can be counted as base
In the extension of the training subset of SVRs, the nested sequence with the increased training subset-SVRs of sample is obtained.In above-mentioned SVR mould
In type, the selection of training subset will occupy a large amount of runing time, and complexity is higher.The prior art three NishidaK, Fujiki
J, KuritaT.Ensemble random-subset SVM.ResearchGate 2016；https://
Www.researchgate.net/publication/288238983. multiple support vector machines are combined, are proposed a kind of new
Integrated random subset SVM algorithm.In the art, the subsample of training set is randomly selected for each SVM, each SVM can be by
It is considered Weak Classifier, there is the random subset SVM being most preferably arranged to obtain by combining all classifiers, but this method is each
It is gradual that the combination of a SVR does not have normal state.Utilities Electric Co. needs accurate prediction address to come price determination and elastic income, into
The scheduling of row energy transfer and unit commitment and sharing of load；The prediction result of low accuracy will cause sizable economic loss,
Prediction error, which increases by 1%, means that power grid and operation cost increase by 10,000,000.The point estimation of electric load have it is uncertain and
The feature of the explanatory difference of confidence level.Therefore an accurate, quick, simple, steady, significant short-term load forecasting mould is developed
Type is very important for Utilities Electric Co. and its client.
In conclusion problem of the existing technology is：The connection between parameter setting and training subset method is not accounted for
System, does not establish the model with U- statistic property and derives its statistical property, can not return to the confidence area of point estimation
Between and confidence level.
Solve the difficulty and meaning of above-mentioned technical problem：Accurate Prediction short-term electric load is Power System Planning and operation
Key.However, the power prediction system based on point estimation will receive the uncertainty and the low limitation of confidence level of itself, and
Establish that the data volume that SVR model is related to is bigger, and computation complexity is very high, precision is general.The present invention is real using sub-sampling strategy
The collateral learning of existing support vector machines (SVR)；Sampling Strategies ensure that each single SVR has enough diversity, and subtract
The loss amount of few information.Then it is selected about model, proposes a kind of new group's Optimization Learning integrated based on single SVR and calculate
Method.Group the advantages of studying in coordination is, it can be ensured that each SVR is integrated, and there is enough intensity to predict short term data.
Integrated approach based on small size sub-sample in the present invention has the statistical inference property of classical U- statistic, mentions
Theoretical guarantee is supplied.The introducing of middle and small scale sub-sample strategy of the present invention, effectively reduces the complexity of estimator, improves simultaneously
Computational accuracy and efficiency, and can return to the confidence interval length and confidence level of preferable electric load point prediction simultaneously, it is
Utilities Electric Co. provides an important risk management tool.
Summary of the invention
In view of the problems of the existing technology, the present invention provides a kind of short term power integrated based on sub-sample SVR is negative
Lotus prediction technique.
The invention is realized in this way the Short-Term Load Forecasting Method integrated based on sub-sample SVR includes：If
Set initial parameter；The model selection and parameter selection of each integrated support vector regression submodel are carried out based on colony intelligence optimization；
Determine the estimation statistic of sub-sample SVR integrated model；Estimate that the SVR Monte Carlo simulation based on sub-sample generates variance；It is defeated
Optimized parameter out, and corresponding sub-sample SVR integrated model.
Step 1 is arranged initial parameter, including collects parameter B on a large scalen, sub-sample size sn。
Step 2 carries out the model selection and parameter choosing of each integrated support vector regression submodel based on colony intelligence optimization
It selects, obtains optimized parameter：
vi(h+1)=rand ()1*vi(h)+c1(h)*rand()2*(pi-xi(h))+c2(h)*rand()3*(pg-xi
(h))；
xi(h+1)=xi(h)+vi(h+1)；
piIt is local optimum position, pgIt is full group's optimal location；rand()1, rand ()2, rand ()3Be be uniformly distributed with
Machine number, c1, c2It is weight parameter；The optimized parameter for generating each sub- integrated model, forms the experts database of parameter.
Step 3, constructs the estimation method of sub-sample SVR integrated model, and estimation statistic is：
Indicate training sample, the estimator constructed in this way has asymptotic normality.
It is given below and generates the process that the sub-sampling support vector regression of asymptotic normality prediction integrates at test point：
Input：Entire training data；
sn：Sub-sample sample size, Bn：Collect parameter on a large scale；
BnA trained subsample；
Output：The final estimation of test sample；BnThe integrated set of individual；
The final estimation of test samplePoint estimation,Variance evaluation；
1) For is recycled, to each positive integer j, 1≤j≤Bn, do parallel；
2) based on j-th of the SVR in j-th of subsample and Share Model selection building；
3) test sample x* is predicted with j-th of SVR；
4)End
5) B is calculatednThe average value of a predicted value is as final estimated value
6) it returns
Step 4 constructs the SVR Monte-Carlo Simulation Method based on sub-sample, and it is as follows to generate variance evaluation：
Specific step is as follows for SVR Monte-Carlo Simulation Method based on sub-sample：
Input：Entire training data；
sn：Sub-sample size, Bn：Collect parameter on a large scale；
BnA trained subsample；
Output：Point estimation,Variance evaluation；
1) to all positive integer i, 1≤i≤BnDo following work
2) s is selected from training datanSubsample Si,j
3) sample S is usediEstablish individual SVR
4) it is predicted using SVR in x*：SVRi(x*)
5)end for
6) average of Monte Carlo simulation is calculated：
7) sample variance is calculated：
8) it returnsWith
In conclusion advantages of the present invention and good effect are：Of support vector machines (SVR) is realized using sampling policy
It practises, ensure that each single SVR has enough diversity, and reduce the loss amount of information, improve computational accuracy and effect
Rate；Group's optimized learning algorithm that model selection is integrated based on single SVR, it is ensured that each SVR is integrated to have enough intensity next pre-
Survey short term data.
There is the statistical inference property of classical U- statistic the present invention is based on the integrated approach of small size sub-sample,
SSVRE mode is easy to be transplanted to parallel computation frame；The confidence level to provide a reference is provided for power scheduling engineer, simultaneously
Also indicate that SSVR model has better and acceptable performance indicator in power-system short-term load forecasting.The present invention constructs more
Simple and higher precision short-term electric load prediction integrated approach, while predicting its point estimation and confidence interval length.
Detailed description of the invention
Fig. 1 is the Short-Term Load Forecasting Method flow chart provided in an embodiment of the present invention integrated based on sub-sample SVR.
Fig. 2 is that the Short-Term Load Forecasting Method provided in an embodiment of the present invention integrated based on sub-sample SVR realizes stream
Cheng Tu.
Fig. 3 is the half an hour electricity in New South Wales provided in an embodiment of the present invention on June 1,1 day to 2007 May in 2007
Power load (measurement unit MW) schematic diagram.
Fig. 4 is the power load of the SSVRE model provided in an embodiment of the present invention based on New South Wales Utilities Electric Co. case
Lotus forecast interval predicts schematic diagram.
Fig. 5 is that the sensitivity analysis of the estimated performance provided in an embodiment of the present invention based on different sub-sampling scales is (integrated
Scale is fixed as 50) schematic diagram.
Fig. 6 is the sensitivity analysis (collection of the forecast confidence provided in an embodiment of the present invention based on different sub-sampling scales
It is fixed as 50) schematic diagram on a large scale.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
The present invention propose a seed sampling support vector regression integrate (SSVRE) come carry out short-term electric load point prediction and
Confidence interval length estimate.
Application principle of the invention is described in detail with reference to the accompanying drawing.
As shown in Figure 1, the Short-Term Load Forecasting Method packet provided in an embodiment of the present invention integrated based on sub-sample SVR
Include following steps：
S101：Initial parameter is set；
S102：The model selection and parameter choosing of each integrated support vector regression submodel are carried out based on colony intelligence optimization
It selects；
S103：Determine the estimation statistic of sub-sample SVR integrated model；
S104：Estimate that the SVR Monte Carlo simulation based on sub-sample generates variance；
S105：Export optimized parameter, and corresponding sub-sample SVR integrated model.
The Short-Term Load Forecasting Method provided in an embodiment of the present invention integrated based on sub-sample SVR specifically includes following
Step：
The first step is arranged initial parameter, including collects parameter B on a large scalen, sub-sample size sn。
Second step, based on colony intelligence optimization carry out each integrated support vector regression (SVR) submodel model selection and
Parameter selection：
vi(h+1)=rand ()1*vi(h)+c1(h)*rand()2*(pi-xi(h))+c2(h)*rand()3*(pg-xi
(h))；
xi(h+1)=xi(h)+vi(h+1)；
piIt is local optimum position, pgIt is full group's optimal location；rand()1, rand ()2, rand ()3Be be uniformly distributed with
Machine number, c1, c2It is weight parameter.The optimized parameter for generating each sub- integrated model, forms the experts database of parameter.
Third step, constructs the estimation method of sub-sample SVR integrated model, and point estimation statistic is：
4th step constructs the SVR Monte-Carlo Simulation Method based on sub-sample, and it is as follows to generate variance evaluation：
Specific step is as follows for SVR Monte-Carlo Simulation Method based on sub-sample：
Input：Entire training data；
sn：Sub-sample size, Bn：Collect parameter on a large scale；
BnA trained subsample；
Output：Point estimation,Variance evaluation；
1) to all positive integer i, 1≤i≤BnDo following work
2) s is selected from training datanSubsample Si,j
3) sample S is usediEstablish individual SVR
4) it is predicted using SVR in x*：SVRi(x*)
5)end for
6) average of Monte Carlo simulation is calculated：
7) sample variance is calculated：
8) it returnsWith
5th step exports optimized parameter, and corresponding sub-sample SVR integrated model.
Application principle of the invention is further described below with reference to specific example.
1, the U- statistic modeling of SVR
According to the definition of U- statistic, it is integrated that the present invention constructs similar sub-sampling support vector regression.It enablesIndicate training sample,It is snThe real value or complex value symmetric function of a variable, one is not
The subsample of complete U- statistic version is integrated to be provided by following formula：
It is the subsample of training sample, sample size Sn,
G is to be estimated by support vector regression model, therefore be denoted as in the present invention：
Since modeling process is unrelated with the sequence of subsample, which is that displacement is symmetrical in its parameter
's.
For a test x*, the integrated estimator of sub-sampling support vector regression is denoted as
When SVR parameter and fixed order, this sub-sampling support vector regression, which integrates estimator, has asymptotic normality.
The present invention is based on the subsample of training set, consider that building SVR is integrated, there are two major advantages:First, what is obtained estimates
Form of the metering using incomplete U statistic, the result of available asymptotic normality；Secondly, it uses the increment of a subset
This, and be easy to determine distribution, reduce computation complexity.
2, the model selection based on group's Optimization Learning
In integrated approach, due to concentrating sampling subsample from identical training data, integrated individual member has phase
As feature, and share the common point that they are executed in parameter space.Under the inspiration of particle group optimizing thought, each
Subsample support vector regression, which integrates (SSVRE), can be seen as particle.On this basis, it proposes a kind of based on group's optimization
The model selection method of study.
The particle group optimizing proposed using Kennedy and Eberhart 1995 for group's Optimization Learning process, the present invention
Algorithm (PSO).Its optimization process is by i-th of position equation xi(h+1) and i-th of rate equation vi(h+1) fixed in moment h
Justice：
vi(h+1)=rand ()1*vi(h)+c1(h)*rand()2*(pi-xi(h))+c2(h)*rand()3*(pg-xi
(h))；
xi(h+1)=xi(h)+vi(h+1)；
piIt is local optimum position, pgIt is full group's optimal location；rand()1, rand ()2, rand ()3Be be uniformly distributed with
Machine number, c1, c2It is weight parameter.
Algorithm 1. is shown in the specific implementation of this part
3, statistical inference
Bagging (bootstrap-aggregating) algorithm classify and return field have the characteristics that it is attracting with
Deep experience performance.A major advantage of bagging is its simplicity, but is difficult to carry out theoretical proof.In order into one
Step development, Bootstrap are replaced by random sub-sampling methods, i.e. subsampling-subbagging scheme.Recently,
Lucas Mentch and Giles Hooker allow confidence it has been proved that the prediction in subbagging is asymptotic normality
Section is with prediction.This is the result is that assuming that slight change occurs for a data point in training sample, individually
What the prediction of device did not obtain under the hypothesis of great changes.The condition make sub-sampling support vector regression it is integrated meet it is asymptotic
The Lindeberg condition of distribution.
Asymptotic normality：WhenAndThen to test point x*, have：
WhereinIt is two estimators with the public sample of cBetween covariance.
Based on above-mentioned asymptotic normality, sub-sampling support vector regression integrated approach proposed by the present invention also has accordingly
Statistical inference.Therefore, the distribution obtained is for point estimation and to establish a kind of outstanding method of forecast interval, by applying it
Generate the preferable confidence interval of distribution.
4, forecast interval
On the basis of asymptotic normality, the present invention carries out point estimation using sub-sample support vector regression estimator, such as
EquationIt is shown.In order to establish confidence interval, present invention needs are estimated
Count the variance in limit normal distribution.For this purpose, the part detailed process is shown in calculation present invention introduces a Monte Carlo estimation method
Method 2.
5, the frame of SSVRE method is summarized as follows：
The first step：Initial parameter is set, including collects parameter B on a large scalen, sub-sample size sn。
Second step：Based on colony intelligence optimization carry out each integrated support vector regression (SVR) submodel model selection and
Parameter selection：
vi(h+1)=rand ()1*vi(h)+c1(h)*rand()2*(pi-xi(h))+c2(h)*rand()3*(pg-xi
(h))；
xi(h+1)=xi(h)+vi(h+1)；
piIt is local optimum position, pgIt is full group's optimal location；rand()1, rand ()2, rand ()3Be be uniformly distributed with
Machine number, c1, c2It is weight parameter.The optimized parameter for generating each sub- integrated model, forms the experts database of parameter.
Third step：The estimation method of sub-sample SVR integrated model is constructed, point estimation statistic is：
4th step constructs the SVR Monte-Carlo Simulation Method based on sub-sample, and it is as follows to generate variance evaluation：
5th step exports optimized parameter, and corresponding sub-sample SVR integrated model.
In order to illustrate more clearly of the present invention, the realization details of each step is shown in algorithm 1 and 2.Then, entirely
The flow chart of algorithm frame is as shown in Figure 2.
Detailed process is as follows for algorithm 1：
Input：Entire training data；
sn：Sub-sample sample size, Bn：Collect parameter on a large scale；
BnA trained subsample；
Output：The final estimation of test sample；BnThe integrated set of individual.
The final estimation of test samplePoint estimation,Variance evaluation；
1.For circulation, to each positive integer j, 1≤j≤Bn, do parallel；
2. based on j-th of the SVR in j-th of subsample and Share Model selection building；
3. being predicted with j-th of SVR test sample x*；
4.End
5. calculating BnThe average value of a predicted value is as final estimated value
6. returning
Detailed process is as follows for algorithm 2：
Input：Entire training data；
sn：Sub-sample size, Bn：Collect parameter on a large scale；
BnA trained subsample；
Output：Point estimation,Variance evaluation；
1.for circulation, to all positive integer i, 1≤i≤BnDo following work；
2. selecting s from training datanSubsample Si,j
3. using sample SiEstablish individual SVR
4. being predicted using SVR in x*：SVRi(x*)
5. terminating for circulation
6. calculating the average of Monte Carlo simulation：
7. calculating sample variance：
8. returningWith
Application effect of the invention is described in detail below with reference to example assessment.
1, model evaluation uses the Power system load data pair in New South Wales county to verify proposed SSVRE model
Its validity is examined.
Load forecast can be divided into long term load forecasting, medium term load forecasting and short-term load forecasting.Wherein, short-term negative
Lotus prediction (such as half an hour electric load) problem will generate a large amount of data in real time, and therefore, the present invention selects short-term load forecasting
To verify model of the invention.In general, Mid-long term load forecasting is mainly referred to by macroeconomy such as economic level, social factors
Target influences, the influence of the short-term cycle indicator such as the main climate factor of short-term load forecasting and itself load factor.For this purpose, this
Invention uses multiple regression and phase space reconfiguration model, verifies for model.
The description of 1.1 data and pretreatment
Electric load unit is megawatt (megawatt) (megawatt be equal to 1000 kilowatts (kilowatt)) in the present invention, and temperature unit is to take the photograph
Family name's degree.
Case New South Wales Utilities Electric Co.
The short-term load forecasting problem (half an hour basis electric load) of New South Wales Utilities Electric Co. includes 1488 half
Hour load data point.In fig. 3 it is shown that the half an hour electricity from the New South Wales on 1 day June 1 to 2007 years Mays in 2007
Power load.
Data prediction：Before carrying out data modeling with statistical method, the normalization technology of original training data is pre-
Survey the key job of design.Most widely used technology is Min-Max scaling and Z-score standardization.During this investigation it turned out,
Data prediction is carried out using the Z score standardization of R software.
It can be the data of zero-mean and unit variance by data zooming according to Z-score standardized technique.Namely
It says, x can be normalized with following equation：
Wherein μ and σ is the mean value and standard variance of raw data set respectively.μ and σ can pass through sample average and sample mark
Quasi- variance statistic amount calculates, and the two parameters should be saved and used for later prediction.In the realization of prediction model
Later, prediction data can be easily restored to not normalized value using following equation.X=x ' × σ+μ
The performance standard of 1.2 predictions
In order to assess the accuracy and confidence level of obtained prognostic experiment, present invention employs three precision evaluations to refer to
Mark and confidence interval length.
Three precision evaluation indexs：
The present invention has selected three kinds of common evaluation indexes, i.e., mean absolute error (MAE), root-mean-square error (RMSE) and
Average absolute percentage error (MAPE).
Here AiIt is i-th of actual data point, PiIt is the data point of prediction, n is the quantity of prediction data point.
Precision of prediction, Er Qieke not only can be improved in confidence interval length, the integrated prediction based on multiple random subsample
To provide its confidence interval length.Confidence interval provides range of indeterminacy, and therefore, confidence interval is longer, uncertain bigger.
Based on this principle, confidence level is predicted in confidence interval Length Indication accordingly, it can provide important wind for Utilities Electric Co.
Dangerous management tool.
1.3 result
Collection is set as fixed value 50 on a large scale, and phase space dimension is arranged to 3.These settings, which can reflect, to be proposed
SSVRE model robustness.
For the actual conditions of New South Wales Utilities Electric Co., as can be seen from Figure 4, relative to original loads data value, 95%
Confidence interval length value it is very small, this show proposed SSVRE prediction framework have very strong robust performance.
The sensitivity analysis of 2SSVRE
Fig. 5 is shown under conditions of size is 50 to fixed set on a large scale, when sub-sampling size is greater than 20, SSVRE mould
Type has good estimated performance, shows the dependable with function of the small sample double sampling strategy of SVR model.Actually answering
In, the complexity of SVR is can be effectively reduced in small sample sub-sampling strategy.
Fig. 6 gives under conditions of size is 50 to fixed set on a large scale, and 95% sets when increasing with sub-sampling size
Believe the change curve of the mean value of siding-to-siding block length.The confidence interval provides uncertainty, i.e. confidence interval is longer, and uncertainty is got over
Greatly.Generally, with the increase of sub-sampling size, the length of confidence interval is all relatively short, this shows that confidence level is higher.?
The data set, the convergence when sub-sample size level reaches 30 of SSVRE model.Therefore, small size sub-sampling strategy is suitble to SVR mould
Type.Based on the above principles, confidence interval length indicates corresponding prediction confidence level, and important wind can be provided for Utilities Electric Co.
Dangerous management tool.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (7)
1. a kind of Short-Term Load Forecasting Method integrated based on sub-sample SVR, which is characterized in that described to be based on sub-sample
SVR integrated Short-Term Load Forecasting Method includes：Initial parameter is set；Each integrated support is carried out based on colony intelligence optimization
The model of vector regression submodel selects and parameter selection；Determine the estimation statistic of sub-sample SVR integrated model；Estimation is based on
The SVR Monte Carlo simulation of sub-sample generates variance；Export optimized parameter and corresponding sub-sample SVR integrated model.
2. the Short-Term Load Forecasting Method integrated as described in claim 1 based on sub-sample SVR, which is characterized in that institute
Stating setting initial parameter includes collection parameter B on a large scalen, sub-sample size sn。
3. the Short-Term Load Forecasting Method integrated as described in claim 1 based on sub-sample SVR, which is characterized in that institute
State the model selection and parameter selection that each integrated support vector regression submodel is carried out based on colony intelligence optimization：
vi(h+1)=rand ()1*vi(h)+c1(h)*rand()2*(pi-xi(h))+c2(h)*rand()3*(pg-xi(h))；
xi(h+1)=xi(h)+vi(h+1)；
Wherein：piIt is local optimum position, pgIt is full group's optimal location；rand()1, rand ()2, rand ()3Be be uniformly distributed with
Machine number, c1, c2It is weight parameter；The optimized parameter for generating each sub- integrated model, forms the experts database of parameter.
4. the Short-Term Load Forecasting Method integrated as described in claim 1 based on sub-sample SVR, which is characterized in that institute
The statistic of estimation method for stating building sub-sample SVR integrated model is：
5. the Short-Term Load Forecasting Method integrated as described in claim 1 based on sub-sample SVR, which is characterized in that institute
It is as follows to state SVR Monte-Carlo Simulation Method generation variance evaluation of the building based on sub-sample：
Specific step is as follows for SVR Monte-Carlo Simulation Method based on sub-sample：
Input：Entire training data；
sn：Sub-sample size, Bn：Collect parameter on a large scale；
BnA trained subsample；
Output：Point estimation,Variance evaluation；
1) for is recycled, to all positive integer i, 1≤i≤BnDo following work；
2) s is selected from training datanSubsample Si,j
3) sample S is usediEstablish individual SVR
4) it is predicted using SVR in x*：SVRi(x*)
5) terminate for circulation
6) average of Monte Carlo simulation is calculated：
7) sample variance is calculated：
8) it returnsWith
6. the Short-Term Load Forecasting Method integrated as described in claim 1 based on sub-sample SVR, which is characterized in that institute
Stating sub-sample SVR integrated model is：
Indicate training sample.
7. the Short-Term Load Forecasting Method integrated as claimed in claim 6 based on sub-sample SVR, which is characterized in that into
One step includes：
Input：Entire training data；
sn：Sub-sample sample size, Bn：Collect parameter on a large scale；
BnA trained subsample；
Output：The final estimation of test sample；BnThe integrated set of individual；
The final estimation of test samplePoint estimation,Variance evaluation；
1) For is recycled, to each positive integer j, 1≤j≤Bn, do parallel；
2) based on j-th of the SVR in j-th of subsample and Share Model selection building；
3) test sample x* is predicted with j-th of SVR；
4)End
5) B is calculatednThe average value of a predicted value is as final estimated value
6) it returns
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810590419.5A CN108879656B (en) | 2018-06-09 | 2018-06-09 | Short-term power load prediction method based on sub-sampling SVR integration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810590419.5A CN108879656B (en) | 2018-06-09 | 2018-06-09 | Short-term power load prediction method based on sub-sampling SVR integration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108879656A true CN108879656A (en) | 2018-11-23 |
| CN108879656B CN108879656B (en) | 2021-08-10 |
Family
ID=64338520
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810590419.5A Active CN108879656B (en) | 2018-06-09 | 2018-06-09 | Short-term power load prediction method based on sub-sampling SVR integration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108879656B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110008626A (en) * | 2019-04-16 | 2019-07-12 | 电子科技大学 | A kind of short term power energy consumption prediction technique based on Spark |
| CN112784477A (en) * | 2021-01-06 | 2021-05-11 | 平衡机器科技(深圳)有限公司 | WRF-LES and BP-PSO-Bagging combined wind power prediction method |
| US20220067258A1 (en) * | 2020-08-26 | 2022-03-03 | Northwestern Polytechnical University | Robust optimal design method for photovoltaic cells |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102360430A (en) * | 2011-09-28 | 2012-02-22 | 哈尔滨工业大学 | Interval forecasting method of heat supply load based on support vector machine and error estimation |
| CN103106329A (en) * | 2012-11-19 | 2013-05-15 | 华北电力大学 | Training sample grouping construction method used for support vector regression (SVR) short-term load forecasting |
| CN107527116A (en) * | 2017-08-28 | 2017-12-29 | 山东科技大学 | short-term load forecasting method based on support vector regression |
-
2018
- 2018-06-09 CN CN201810590419.5A patent/CN108879656B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102360430A (en) * | 2011-09-28 | 2012-02-22 | 哈尔滨工业大学 | Interval forecasting method of heat supply load based on support vector machine and error estimation |
| CN103106329A (en) * | 2012-11-19 | 2013-05-15 | 华北电力大学 | Training sample grouping construction method used for support vector regression (SVR) short-term load forecasting |
| CN107527116A (en) * | 2017-08-28 | 2017-12-29 | 山东科技大学 | short-term load forecasting method based on support vector regression |
Non-Patent Citations (3)
| Title |
|---|
| JINXING CHE等: "Optimal training subset in a support vector regression electric load forecasting model", 《APPLIED SOFT COMPUTING》 * |
| MIN ZOU等: "Load Forecast Based on Least Square SVR and HHT", 《2017 13TH INTERNATIONAL CONFERENCE ON NATURAL COMPUTATION, FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY》 * |
| YANYING LI等: "A modified support vector regression Integrated selection of training subset and model", 《APPLIED SOFT COMPUTING》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110008626A (en) * | 2019-04-16 | 2019-07-12 | 电子科技大学 | A kind of short term power energy consumption prediction technique based on Spark |
| US20220067258A1 (en) * | 2020-08-26 | 2022-03-03 | Northwestern Polytechnical University | Robust optimal design method for photovoltaic cells |
| US11630936B2 (en) * | 2020-08-26 | 2023-04-18 | Northwestern Polytechnical University | Robust optimal design method for photovoltaic cells |
| CN112784477A (en) * | 2021-01-06 | 2021-05-11 | 平衡机器科技(深圳)有限公司 | WRF-LES and BP-PSO-Bagging combined wind power prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108879656B (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Local energy trading behavior modeling with deep reinforcement learning | |
| Wang et al. | Synchronous pattern matching principle-based residential demand response baseline estimation: Mechanism analysis and approach description | |
| Su et al. | A new hybrid model optimized by an intelligent optimization algorithm for wind speed forecasting | |
| Yuan et al. | Short-term wind power prediction based on LSSVM–GSA model | |
| Nie et al. | Hybrid of ARIMA and SVMs for short-term load forecasting | |
| Li et al. | Data-driven charging strategy of PEVs under transformer aging risk | |
| Dong et al. | Storage sizing with peak-shaving policy for wind farm based on cyclic Markov chain model | |
| Morsalin et al. | Electric vehicle charge scheduling using an artificial neural network | |
| CN106251027B (en) | Electric load probability density Forecasting Methodology based on fuzzy support vector quantile estimate | |
| CN108879656A (en) | A kind of Short-Term Load Forecasting Method integrated based on sub-sample SVR | |
| Liu et al. | Distribution system state estimation with high penetration of demand response enabled loads | |
| Luo et al. | Optimal operation scheduling for microgrid with high penetrations of solar power and thermostatically controlled loads | |
| Pan et al. | Analyzing big smart metering data towards differentiated user services: A sublinear approach | |
| Lu et al. | Effective end-to-end learning framework for economic dispatch | |
| Apribowo et al. | Optimal planning of battery energy storage systems by considering battery degradation due to ambient temperature: A review, challenges, and new perspective | |
| Das et al. | A novel fitted rolling horizon control approach for real-time policy making in microgrid | |
| Zhang et al. | Transfer learning featured short-term combining forecasting model for residential loads with small sample sets | |
| Hosseini et al. | Data-driven multi-resolution probabilistic energy and reserve bidding of wind power | |
| Hekmat et al. | Data-driven demand-side flexibility quantification: Prediction and approximation of flexibility envelopes | |
| Souhe et al. | Forecasting of electrical energy consumption of households in a smart grid | |
| Kramer et al. | Power prediction in smart grids with evolutionary local kernel regression | |
| Kazhamiaka et al. | Comparison of different approaches for solar PV and storage sizing | |
| Khosravi et al. | Wind farm power uncertainty quantification using a mean-variance estimation method | |
| Jaramillo et al. | Distributed energy resources electric profile identification in low voltage networks using supervised machine learning techniques | |
| Jiang et al. | Two‐level area‐load modelling for OPF of power system using reinforcement learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |