CN106611243A - Residual correction method for wind speed prediction based on GARCH (Generalized ARCH) model - Google Patents

Abstract

The invention discloses a residual correction method for wind speed prediction based on a GARCH (Generalized ARCH) model, belongs to the technical field of power transmission and distribution, and aims to improve an original wind speed prediction method and improve the wind speed prediction precision. The technical scheme is as follows: adopting the GARCH model to predict an original wind speed series; establishing a regression model for a residual fitted by the GARCH model, and using the regression model to predict the following residuals; and finally, using a predicted value to correct a preliminary prediction result of the GARCH model. The experimental result shows that the residual correction method for the wind speed prediction based on the GARCH model provided by the invention is better than a prediction method of a traditional ARMA and a prediction method of a BP neural network model, and has higher prediction accuracy.

Description

A kind of forecasting wind speed residual GM method based on GARCH models

Technical field

The present invention relates to it is a kind of can accurately prediction of wind speed size so that predict wind power method, belong to power transmission and distribution A kind of technical field, and in particular to forecasting wind speed residual GM method based on GARCH models.

Background technology

The energy is that various countries develop one of indispensable fundamental, and with the fast development of global economy, to the energy Demand also increasingly increase.But traditional fossil energy reserves are limited, with non-renewable, once exhausted human development will Face great crises；The environmentally harmful gases such as another aspect fossil energy burning release SO2, can also discharge carbon dioxide etc. Greenhouse gases cause global warming, therefore various countries will appreciate that it is the mankind to break away from the dependence development new forms of energy to fossil energy The necessary choice of sustainable development.

Through decades development, wind-power electricity generation become that in the world generally acknowledged technology is most ripe, development cost is minimum, most One of regenerative resource of development prospect, is increasingly becoming the object that various countries develop emphatically.In recent years in order to consolidate the development of wind-powered electricity generation, The competitiveness of wind-powered electricity generation is improved, various countries each department all suffer from a series of transformation of policies.Navigant research issue report Claim, following 5～6 years, the investment should not be underestimated is become the most representational energy industry in the whole world by wind energy.2000-2015 Year, global installed capacity of wind-driven power develops into 432.9GW from 17.4GW, increases about 23 times.The newly-increased installation of domestic wind-powered electricity generation is held within 2015 Amount reaches 30.5GW, creates new peak and the continuous 6 years whole world of taking the lead in race.According to the planning of National Energy Board, national wind-powered electricity generation is opened within 2016 Send out and build total scale 30.8GW, the scale of slightly above 2015, it is contemplated that wind-powered electricity generation industry will still maintain the high degree of prosperity.But due to wind Speed has intermittent and undulatory property so that jumbo wind power integration electrical network can to the safe and stable operation of power system and Ensure that the quality of power supply brings severe challenge.Therefore, accurately the following wind speed size of prediction is counted to the scheduling of power department reasonable arrangement Draw, the impact damage tool that reduction moment extreme wind speed is caused to generating set is of great significance, and to a certain extent The further development of wind-powered electricity generation is promoted, the market competitiveness of wind-powered electricity generation is improved.

The content of the invention

The technical problem to be solved is to provide a kind of forecasting wind speed residual GM side based on GARCH models Method, it is intended in order to reduce wind-abandoning phenomenon for the balance that disappears, obtain accurate prediction of wind speed most important.

To solve above-mentioned technical problem, the technical solution used in the present invention is：A kind of wind speed based on GARCH models is pre- Survey residual GM method, it is characterised in that：

Wind energy turbine set historical wind speed data to be predicted are collected, and its pretreatment is fitted to into historical wind speed sequence y_t-1, then will Historical wind speed sequence y_t-1It is fitted to GARCH models and wind speed is predicted, the wind for obtaining will be predicted based on time serieses Fast prediction data y'_tIt is fitted to prediction of wind speed sequenceAccording to historical wind speed sequence y_t-1With prediction of wind speed sequenceObtain The regression criterion sequence of GARCH modelsAnd to the regression criterion sequence of GARCH modelsRegression model is set up, is predicted The residual sequence predictive value of GARCH modelsFinally with the regression criterion sequence prediction value of GARCH modelsAmendment GARCH moulds The predicting the outcome of type obtains wind energy turbine set to be predicted and predicts the outcome correction valueI.e.

Further technical scheme is, the wind energy turbine set to be predicted for obtaining is predicted the outcome correction valueAs forecasting wind speed Data y'_t, using the air speed data pretreatment of wind energy turbine set t time to be predicted as historical wind speed sequence a part, and then obtain t The forecasting wind speed modified result value of+1 time

Further technical scheme is that the GARCH models are

y_t=x_tβ+ε_t (1)

Residual error stochastic process { ε_tObey following process：

Wherein, α_jAnd φ_iFor unknown parameter；

It is above-mentioned, { ε_tIt is generilized auto regressive conditional heteroskedastic process, it is designated as { ε_t}～GARCH (p, q).

Further technical scheme is, the variance yields h of the choosing value of p, q in the GARCH (p, q) according to the front moment_t With the variance yields h of its previous moment_t-1And residual values ε_t-1Contact compactness is selected；

Wherein p=1；Q=1.

Further technical scheme is also resided in, and regression model is set up to the regression criterion sequence of GARCH models, wherein returning The exponent number AIC criterion of model is defined below：

It is using the beneficial effect produced by above-mentioned technical proposal：What the present invention was selected is based on the time of statistical method Sequence method.In classical time serieses field, linear model, in occupation of consequence, is the basis of other models.Linear mould Type simple structure, model parameter estimation and Forecasting Methodology it is more ripe, significant in time series analysis and status.This Invention is contrasted with the present extensive arma modeling of comparison used and BP neural network model, respectively to Spain's wind energy turbine set with Gansu, China wind farm data is modeled analysis, homogeneous using mean absolute error (MAE), root-mean-square error (RMSE) peace Comprehensive assessment is carried out to result to three error criterions of error (MRE) and the simulation curve that predicts the outcome, and result is compared： Either every evaluation index of the present invention is superior to the time to be predicted to Spain's wind farm data or Gansu wind farm data Sequential forecasting models BP models and GARCH (1,1) forecast model of residual GM is not carried out.

Description of the drawings

With reference to the accompanying drawings and detailed description the present invention is further detailed explanation.

Fig. 1 is the modeling data of model；

Wherein, (a) for Spain's wind energy turbine set air speed data；B () is the modeling data of Gansu wind energy turbine set；

Fig. 2 is the actual value and predictive value of Spain's wind farm wind velocity data

Wherein, (a) be GARCH models based on residual GM and arma modeling comparison diagram；B () is based on residual GM GARCH models and BP models comparison diagram；

Fig. 3 is the actual value and predictive value of Gansu wind farm wind velocity data

Wherein, (a) be GARCH models based on residual GM and arma modeling comparison diagram；B () is based on residual GM GARCH models and BP models comparison diagram；

Fig. 4 is the amendment flow process of the present invention of embodiment 2；

Fig. 5 is the amendment flow process of the present invention of embodiment 3.

Specific embodiment

With reference to the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Ground description, it is clear that described embodiment a part of embodiment only of the invention, rather than the embodiment of whole.It is based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made Embodiment, belongs to the scope of protection of the invention.

Many details are elaborated in the following description in order to fully understand the present invention, but the present invention can be with It is different from alternate manner described here to implement using other, those skilled in the art can be without prejudice to intension of the present invention In the case of do similar popularization, therefore the present invention is not limited by following public specific embodiment.

The invention provides a kind of forecasting wind speed residual GM method based on GARCH models, here, ARCH models most base This be characterised by it is assumed that the error term of a substantially linear regression model conditional variance present autocorrelation performance, it when Between sequence dynamic model promoted, the time dependent feature of conditional variance of disturbance term is described with ARCH models.Due to ARCH models deficiency in practice, by model further genralrlization to general ARCH models, i.e. GARCH models.GARCH models ratio ARCH models need less lag order.The residual error after GARCH models fittings is built to improve the forecasting wind speed precision present invention Vertical regression model, and by the regression model prediction of residual error later, finally with the predictive value to GARCH model preliminary forecastings Result be modified.Methods described comprises the steps：

A. the historical wind speed data of wind energy turbine set to be predicted are collected；

In order to obtain following wind series predictive value sometime, the data that need to be collected can not only include historical wind speed number According to also including forecasting wind speed data；

B. historical wind speed data are carried out with historical wind speed sequence y that pretreatment operation forms actual modeling_t-1；

Can checking sequence stationarity, judge nonlinear characteristic of data sequence etc..

C. to historical wind speed sequence y_t-1General autoregressive conditional different Variance model is set up, i.e. (GARCH) model, fitting is simultaneously Forecasting wind speed data y'_t；

It can be seen from the seriality principle of time, the variance yields h at current time_tWith the variance yields ht of its previous moment_-1And Residual valuesThe most closely, therefore Conditional heterosedasticity model may be selected GARCH (1,1) model, i.e. exponent number be 1 for contact；

D. regression model is set up to the regression criterion of GARCH models, the exponent number AIC criterion of wherein regression model is as follows：

If working as n=n₀When, AIC (n, n-1)=min then shows that applicable model of fit is ARMA (n₀,n₀-1)；

E. with the regression criterion sequence prediction value of GARCH modelsThat is, the predictive value of regression model in d, corrects GARCH Predicting the outcome for model is obtainedI.e.

Wherein, GARCH model modelings are as follows：

Wherein the most basic feature of ARCH models is it is assumed that the condition of the error term of a substantially linear regression model Variance is presented autocorrelation performance.ARCH (p) models that Engle is proposed are as follows：

y_t=x_tβ+ε_t (1)

If residual error stochastic process { ε_tThe following process of obedience, i.e.,

Wherein, α_jFor unknown parameter.In order to guaranteed conditions variance is positive number, it is desirable to α_j>=0, j=0,1 ..., q；In order to protect Card { ε_tSteady, it is desirable to α₁+α₂+…+α_q＜ 1.

Due to ARCH models deficiency in practice, when p is larger, parameter estimation is no longer accurate, it is impossible to ensure variance for just. Bollerslev is by model further genralrlization to general ARCH (Generalized ARCH, GARCH) model.GARCH models ratio ARCH models need less lag order.If GARCH is model residual error stochastic process { ε_tObey following process, i.e.,

Wherein, α_jAnd φ_iFor unknown parameter.Then claim { ε_tIt is generilized auto regressive conditional heteroskedastic process, it is designated as { ε_t}～ GARCH(p,q)。

Understood according to the seriality principle and formula (3) of time, the variance yields h at current time_tWith the variance of its previous moment Value h_t-1And residual values ε_t-1Contact is the tightst, thus selection GARCH (1,1) model.Finally, to above AR (1)-GARCH (1, 1) regression criterion of model sets up regression model, inspection autocorrelation of residuals and partial correlation, determines back in combination with AIC criterion Return the optimal fitting exponent number of model.

The present invention proposes a kind of method of the forecasting wind speed of the GARCH models based on residual GM, institute's extracting method with it is existing With the extensive arma modeling of comparison and BP neural network model contrasted.The present invention is carried out in detail with reference to example Explanation：

Step one：Have chosen domestic and international two wind farm datas to be verified.One be Spain's wind energy turbine set wind speed number According to totally 150 data, wherein 100 sample datas, 50 data to be predicted；Another is Gansu, China wind farm wind velocity number According to totally 300 data, wherein 250 sample datas, 50 data to be predicted.

Step 2：All data to collecting carry out pretreatment operation, including remove thick value etc..

Step 3：GARCH is set up to sample data, and (1,1) model is fitted sample data and calculates next step predictive value y'_t Then it is fitted to prediction of wind speed sequence

Step 4：Regression model, the next step predictive value of digital simulation residual error are set up using regression criterion

Step 5：Predicted the outcome using the prediction wind energy turbine set of the predictive value amendment GARCH models of regression model in step 4 Correction valueI.e.

Interpretation

The present invention is by Spain's wind farm data and Gansu, China wind farm wind velocity data to proposed by the invention Method carry out case verification, accompanying drawing illustrates the main experimental results of the present invention.Illustrate hereby, following experimental analysiss are only to show Model, rather than the method is confined in specific application environment.This paper institute's extracting methods and the present extensive ARMA moulds of comparison Type and BP neural network model are contrasted, and are modeled analysis to Spain's wind energy turbine set and Gansu wind farm data respectively, should With mean absolute error (MAE), three error criterions of root-mean-square error (RMSE) and average relative error (MRE) and predict the outcome Simulation curve carries out comprehensive assessment to result, and the result of calculation of forecasting wind speed error criterion is as shown in table 2.

The Algorithm Error index of table 2 is contrasted

By analytical table 2, either Spain's wind farm data or Gansu wind farm data are predicted, based on residual The forecasting wind speed model items evaluation index of the GARCH of difference amendment is superior to time series predicting model and does not carry out residual error to repair Positive GARCH (1,1) forecast model.Error correction GARCH model is set up to Spain and Gansu wind farm data, with ARMA moulds Type compares mean absolute error and have dropped 53.87% and 64.09% respectively, and root-mean-square error have dropped 47.88% He 67.79%, average relative error have dropped 58.57% and 69.47%.Mean absolute error have dropped respectively compared with BP models 52.52% and 60.43%, root-mean-square error have dropped 43.15% and 58.81%, and average relative error have dropped 57.73% He 61.01%.With GARCH (1,1) model compare mean absolute error and have dropped 47.81% and 59.98%, root-mean-square error declines 43.33% and 60.01%, mean absolute error have dropped 49.79% and 59.96%.Known by Fig. 2, due to Gansu wind energy turbine set Data fluctuations scope is less, relatively steady, therefore the precision of prediction of each model is higher compared with Spain's wind energy turbine set.In summary, herein The forecasting wind speed model prediction accuracy of the GARCH based on residual GM for proposing is significantly larger than simple traditional forecasting wind speed mould Type, with actual application value.

Claims (5)

1. a kind of forecasting wind speed residual GM method based on GARCH models, it is characterised in that：

Wind energy turbine set historical wind speed data to be predicted are collected, and its pretreatment is fitted to, then by historical wind speed sequence y_t-1Fitting It is predicted for GARCH models and to wind speed, forecasting wind speed data y' for obtaining will be predicted based on time serieses_tIt is fitted to Prediction of wind speed sequenceAccording to historical wind speed sequence y_t-1With prediction of wind speed sequenceObtain the regression criterion sequence of GARCH modelsAnd to the regression criterion sequence of GARCH modelsRegression model is set up, the residual sequence predictive value of GARCH models is predictedFinally with the regression criterion sequence prediction value of GARCH modelsPredicting the outcome for GARCH models of amendment obtains wind-powered electricity generation to be predicted Field prediction modified result valueI.e.

2. a kind of forecasting wind speed residual GM method based on GARCH models according to claim 1, it is characterised in that： The wind energy turbine set to be predicted for obtaining is predicted the outcome correction valueAs forecasting wind speed data y'_t, by the wind energy turbine set t time to be predicted Air speed data pretreatment as historical wind speed sequence a part, and then obtain the t+1 times forecasting wind speed modified result value

3. a kind of forecasting wind speed residual GM method based on GARCH models according to claim 1, it is characterised in that： The GARCH models are

y_t=x_tβ+ε_t (1)

Residual error stochastic process { ε_tObey following process：

$\left\{\begin{array}{c}{\mathrm{\ϵ}}_t=\sqrt{h_t}{v}_t\\ h_t={\mathrm{\α}}_0+\underset{i=1}{\overset{p}{\Σ}}{\mathrm{\φ}}_i{h}_{t-q}+\underset{j=1}{\overset{q}{\Σ}}{\mathrm{\α}}_j{\mathrm{\ϵ}}_{t-j}^2\\ v_t~N(0,1)i.i.d\end{array}\right.---\left(3\right)$

Wherein, α_jAnd φ_iFor unknown parameter；

It is above-mentioned, { ε_tIt is generilized auto regressive conditional heteroskedastic process, it is designated as { ε_t}～GARCH (p, q).

4. a kind of forecasting wind speed residual GM method based on GARCH models according to claim 3, it is characterised in that： Variance yields h of the choosing value of p, q in the GARCH (p, q) according to the front moment_tWith the variance yields h of its previous moment_t-1And residual error Value ε_t-1Contact compactness is selected；

Wherein p=1；Q=1.

5. a kind of forecasting wind speed residual GM method based on GARCH models according to claim 1, it is characterised in that： Regression model is set up to the regression criterion sequence of GARCH models, the exponent number AIC criterion of wherein regression model is defined below：

$AIC(n,n-1)=log\widehat{{\mathrm{\σ}}^2}+\frac{4n}{N}$