CN103473322A - Photovoltaic generation power ultra-short term prediction method based on time series model - Google Patents

Abstract

The invention discloses a photovoltaic generation power ultra-short term prediction method based on a time series model. The method is characterized by comprising the following steps of collecting and normalizing historic power data of a photovoltaic power station; establishing a fitting equation according to the normalized historic power data, and determining the order of a model according to the established fitting equation and residual variance, namely, the values of p and q; determining the value of A; establishing an auto-regressive moving average model. According to the photovoltaic generation power ultra-short term prediction method based on the time series model, by establishing the prediction model for the ultra-short term prediction of photovoltaic generation power through the historic power data, the fitting equation and the auto-regressive moving average model, so that the aim of short-term accurate prediction of the photovoltaic generation power can be achieved according to the exiting model and the existing data.

Description

The photovoltaic generation power ultra-short term Forecasting Methodology of time-based series model

Technical field

The present invention relates to field of photovoltaic power generation, particularly, relate to a kind of photovoltaic generation power ultra-short term Forecasting Methodology of time-based series model.

Background technology

Solar energy resources is abundant, widely distributed, is the most potential regenerative resource of 21 century.Along with the problems such as global energy shortage and environmental pollution become increasingly conspicuous, solar energy power generating because it cleans, safety, facility, the characteristics such as efficient, the new industry that has become the countries in the world common concern and given priority to." 12 " China in period will increase approximately 1,000 ten thousand kilowatts of solar photovoltaic power plant installed capacitys newly, 1,000,000 kilowatts of solar light-heat power-generation installed capacitys, approximately 1,000 ten thousand kilowatts of distributed photovoltaic power generation systems.In the more than ten years from now on, the market of Chinese photovoltaic generation will enter the high-speed developing period.

By in August, 2013, the installed capacity of Gansu Power Grid grid-connected photovoltaic surpasses 1,400,000 kilowatts, becomes the second largest photovoltaic generation base that is only second to Qinghai.Grid-connected on a large scale along with photovoltaic plant, uncertainty and uncontrollability that photovoltaic generation is exerted oneself are brought problems to Regulation,

Therefore, when grid-connected capacity scale is larger, develop practical photovoltaic generation power ultra-short term prognoses system, can effectively reduce spinning reserve capacity, improve the power grid security economic operation level.

Denmark, Germany, Italy, Spain, the U.S., Japan and other countries have all been carried out the correlative study of photovoltaic generation power forecasting method, and Countries has formed relevant product and obtained the scale application simultaneously.The research of photovoltaic generation power forecasting method is at the early-stage at home, some Patents for the photovoltaic generation power prediction have appearred, comprise utilize expert knowledge library carry out the photovoltaic generation power prediction method, utilize the BP neural network to carry out method of photovoltaic generation power prediction etc.But the photovoltaic generation power ultra-short term Forecasting Methodology that the time-based series model not yet occurs.

Summary of the invention

The object of the invention is to, for the problems referred to above, propose a kind of photovoltaic generation power ultra-short term Forecasting Methodology of time-based series model, to realize photovoltaic generation power is carried out the advantage of short-term Accurate Prediction.

For achieving the above object, the technical solution used in the present invention is:

A kind of photovoltaic generation power ultra-short term Forecasting Methodology of time-based series model comprises the following steps:

Step 1: gather the historical power data of photovoltaic plant, and historical power data is carried out to normalized, the normalized formula is as follows:

$x_i^{\′}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}$

Wherein, x _maxthe maximal value in historical power data, x _minthe minimum value in historical power data, x _ii historical power data, x ' _ii historical power data after normalization;

Step 2: set up fit equation according to historical power data after above-mentioned normalization, according to fit equation and the residual error variance set up, determine that the exponent number of model is the value of p and q;

Step 3: power data as historical as the photovoltaic plant gathered in step 1 utilizes data sequence x ₁, x ₂..., x _tmean, its sample autocovariance is defined as

${\widehat{\mathrm{\γ}}}_k=\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k},k=\mathrm{0,1,2},...,n-1,$

Wherein

x _tand x _t-kbe data sequence x ₁, x ₂..., x _tin numerical value;

Historical power data sample autocorrelation function is:

${\widehat{\mathrm{\ρ}}}_k=\frac{{\widehat{\mathrm{\γ}}}_k}{{\widehat{\mathrm{\γ}}}_0}=\frac{\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\frac{1}{n}\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2}=\frac{\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2},$

$k=\mathrm{0,1,2},...,n-1$

The square of autoregression part is estimated as

Order

Covariance function is

With

estimation replace γ _k,

Draw,

To moving average model coefficient θ ₁, θ ₂..., θ _qthe employing square is estimated, is had

${\mathrm{\γ}}_0=(1+{\mathrm{\θ}}_1^2+{\mathrm{\θ}}_2^2+...+{\mathrm{\θ}}_q^2){\mathrm{\σ}}_a^2$

${\mathrm{\γ}}_k=(-{\mathrm{\θ}}_k+{\mathrm{\θ}}_1+{\mathrm{\θ}}_{k+1}+...+{\mathrm{\θ}}_{q-k}{\mathrm{\θ}}_q){\mathrm{\σ}}_a^2,$

$k=\mathrm{1,2},...,m$

Below comprise altogether m+1 equation, for its parameter, equation is non-linear, adopts process of iteration to be solved and draws θ ₁, θ ₂..., θ _qvalue;

Step 4: the p drawn according to above-mentioned steps two and step 3 and q value and

θ ₁, θ ₂..., θ _qvalue set up autoregressive moving-average model;

The formula of described autoregressive moving-average model is as follows:

Wherein, and θ _j(1≤j≤q) is coefficient, α _tit is white noise sequence.

According to a preferred embodiment of the invention, in above-mentioned steps two, set up fit equation, the exponent number of determining model according to the fit equation of setting up and residual error variance is: the model increased progressively gradually with serial exponent number carrys out the matching original series, all calculates residual sum of squares (RSS) at every turn then draw exponent number and

figure, when exponent number during by little increase,

can significantly descend, after reaching true exponent number

value tend towards stability gradually, the estimator of residual error variance is:

The observed value item number of " actual observed value number " actual use while referring to model of fit, for the sequence with N observed value, the matching autoregressive model, the actual observed value of using mostly is N-p most.

According to a preferred embodiment of the invention, above-mentioned matching autoregressive model formula is as follows:

Coefficient wherein

be called autoregressive coefficient, α _tmean residual sequence, and meet E (α _t)=0, α _tfor white noise sequence, { X _t, t=... ,-2 ,-1,0,1,2 ... mean the observed reading of random series, X _t-1, X _t-2..., X _t-pfor front p observed reading constantly.

Technical scheme of the present invention has following beneficial effect:

Technical scheme of the present invention by historical power data and fit equation and autoregressive moving-average model equation to photovoltaic generation power ultra-short term prediction set up forecast model, can reach the purpose of photovoltaic generation power being carried out to the short-term Accurate Prediction according to forecast model and existing data.

Below by drawings and Examples, technical scheme of the present invention is described in further detail.

The accompanying drawing explanation

The photovoltaic generation power ultra-short term Forecasting Methodology process flow diagram that Fig. 1 is the described time-based series model of the embodiment of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are described, should be appreciated that preferred embodiment described herein, only for description and interpretation the present invention, is not intended to limit the present invention.

As shown in Figure 1, a kind of photovoltaic generation power ultra-short term Forecasting Methodology of time-based series model comprises the following steps:

Step 101: gather the historical power data of photovoltaic plant, and historical power data is carried out to normalized, the normalized formula is as follows:

$x_i^{\′}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}$

Wherein, x _maxthe maximal value in historical power data, x _minthe minimum value in historical power data, x _ii historical power data, x _ii historical power data after normalization;

Step 102: set up fit equation according to historical power data after above-mentioned normalization, according to fit equation and the residual error variance set up, determine that the exponent number of model is the value of p and q;

Step 103: power data as historical as the photovoltaic plant gathered in step 1 utilizes data sequence x ₁, x ₂..., x _tmean, its sample autocovariance is defined as

${\widehat{\mathrm{\γ}}}_k=\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k},k=\mathrm{0,1,2},...,n-1,$

Wherein

x _tand x _t-kbe data sequence x ₁, x ₂..., x _tin numerical value;

Historical power data sample autocorrelation function is:

${\widehat{\mathrm{\ρ}}}_k=\frac{{\widehat{\mathrm{\γ}}}_k}{{\widehat{\mathrm{\γ}}}_0}=\frac{\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\frac{1}{n}\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2}=\frac{\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2},$

$k=\mathrm{0,1,2},...,n-1$

The square of autoregression part is estimated as

Order

Covariance function is

With

estimation replace γ _k,

Draw,

To moving average model coefficient θ ₁, θ ₂..., θ _qthe employing square is estimated, is had

${\mathrm{\γ}}_0=(1+{\mathrm{\θ}}_1^2+{\mathrm{\θ}}_2^2+...+{\mathrm{\θ}}_q^2){\mathrm{\σ}}_a^2$

${\mathrm{\γ}}_k=(-{\mathrm{\θ}}_k+{\mathrm{\θ}}_1{\mathrm{\θ}}_{k+1}+...+{\mathrm{\θ}}_{q-k}{\mathrm{\θ}}_q){\mathrm{\σ}}_a^2,$

$k=\mathrm{1,2},...,m$

Below comprise altogether m+1 equation, for its parameter, equation is non-linear, adopts process of iteration to be solved and draws θ ₁, θ ₂..., θ _qvalue;

Step 104: the p drawn according to above-mentioned steps two and step 3 and q value and

θ ₁, θ ₂..., θ _qvalue set up autoregressive moving-average model;

The formula of described autoregressive moving-average model is as follows:

Wherein,

and θ _j(1≤j≤q) is coefficient, α _tit is white noise sequence.

According to a preferred embodiment of the invention, in above-mentioned steps two, set up fit equation, the exponent number of determining model according to the fit equation of setting up and residual error variance is: the model increased progressively gradually with serial exponent number carrys out the matching original series, all calculates residual sum of squares (RSS) at every turn then draw exponent number and

figure, when exponent number during by little increase,

can significantly descend, after reaching true exponent number

value tend towards stability gradually, the estimator of residual error variance is:

The observed value item number of " actual observed value number " actual use while referring to model of fit, for the sequence with N observed value, the matching autoregressive model, the actual observed value of using mostly is N-p most.

According to a preferred embodiment of the invention, above-mentioned matching autoregressive model formula is as follows:

Coefficient wherein

be called autoregressive coefficient, α _tmean residual sequence, and meet E (α _t)=0, α _tfor white noise sequence, { X _t, t=... ,-2 ,-1,0,1,2 ... mean the observed reading of random series, X _t-1, X _t-2..., X _t-pfor front p observed reading constantly.

The random time series model can be divided into: autoregressive model (AR), moving average model (MA), autoregressive moving-average model (ARMA).

With { X _t, t=... ,-2 ,-1,0,1,2 ... mean the observed reading of random series, X _twith front p observed reading X constantly _t-1, X _t-2..., X _t-pcorrelativity or dependence are arranged, the random difference equation of useable linear regression model:

Describe, be designated as AR (p), wherein coefficient

be called autoregressive coefficient, α _tmean residual sequence, and meet E (α _t)=0, separate, and variance is

usually claim α _tfor white noise sequence.

As current time is t, and stationary random sequence X _tobserved reading before moment t is, X _t-1, X _t-2..., X _t-p, now utilize sequence X _tsequential value to t after constantly predicted, this prediction is called take t as initial point, the prediction that step-length is 1, and predicted value is designated as

x _t+lone-step prediction be:

In reality, application is moving average model MA widely, and q rank moving average model is designated as MA (q), means time series X _tat t observed reading constantly only with white noise sequence α _t-1, α _t-2..., α _t-qrelevant, and with more early stage sequence α _t-j(j=q+1, q+2 ... .) irrelevant.Q rank moving average model can be formulated as:

$X_t={\mathrm{\α}}_t-{\mathrm{\θ}}_1{\mathrm{\α}}_{t-1}-{\mathrm{\θ}}_2{\mathrm{\α}}_{t-2}-...-{\mathrm{\θ}}_q{\mathrm{\α}}_{t-q}=\underset{t=0}{\overset{q}{\mathrm{\Σ}}}{\mathrm{\θ}}_i{\mathrm{\α}}_{t-i}---\left(3\right)$

Wherein, α _tfor white noise sequence, E (α _t) ²=σ ², θ _j(1≤j≤q) is weight, and θ=(θ ₁, θ ₂..., θ _q) meet reversal condition.The variance of 1 step lowest mean square root error is only relevant with prediction step 1, and irrelevant with the timeorigin t of prediction, the steady character that this predict, and when prediction step is larger, the variance of predicated error is also larger, and the accuracy of prediction is lower.

Autoregressive moving-average model ARMA (p, q) combines above-mentioned two kinds of models, and the mixture model of a kind of p rank autoregression q rank running mean obtained, mathematical description is

Wherein, and θ _j(1≤j≤q) is coefficient, α _tit is the white noise sequence met the demands.

The advantage of time series forecasting algorithm is: calculated amount is smaller, computing velocity is very fast, to length of history data, require lower; The deficiency of time series forecasting algorithm is: reflection meteorologic factor that can not be responsive, exert oneself for change the photovoltaic generation cause because of meteorology that change can not Accurate Prediction, higher to the historical data accuracy requirement, the impact that is subject to singular data of predicting the outcome is very large, need the strict historical data of processing, guarantee data accuracy, along with the increase precision of prediction of prediction step is more and more lower.Therefore, the time series forecasting algorithm is suitable for ultra-short term photovoltaic generation power prediction.

The specific algorithm implementation step is as follows:

1, training data normalized

The historical power data of photovoltaic plant is handled as follows,

$x_i^{\′}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}$

2: model is determined rank

Need to set up estimation function with the item of how many known time sequences owing to can't determining in advance, so need to carry out determining the rank judgement to model.

If x _tfor the item that needs are estimated, x _t-1, x _t-2..., x _t-nfor known historical power sequence, for ARMA (p, q) model, it is exactly the value of determining Model Parameter p and q that model is determined rank.

In technical scheme of the present invention, adopt residual error variogram method to carry out model and determine rank.Hypothetical model is limited rank autoregressive models, if the exponent number arranged is less than true exponent number, be a kind of not enough matching, thereby the matching residual sum of squares (RSS) must be bigger than normal, now by improving exponent number, can significantly reduce residual sum of squares (RSS).Otherwise, if exponent number has reached actual value, increase so again exponent number, be exactly overfitting, now increase exponent number and can not make residual sum of squares (RSS) significantly reduce, even can slightly increase.

The model increased progressively gradually with a series of exponent numbers like this carrys out the matching original series, all calculates residual sum of squares (RSS) at every turn then draw exponent number and

figure.When exponent number during by little increase,

can significantly descend, after reaching true exponent number

value can tend towards stability gradually.The estimator of residual error variance is:

The observed value item number of " actual observed value number " actual use while referring to model of fit, for the sequence with N observed value, matching AR (p) model, the actual observed value of using mostly is N-p most.

" model parameter number " refers to the number of parameters that in set up model, actual packet contains, and for the model that contains average, the model parameter number is that model order adds 1.For the sequence of N observed reading, the residual error estimator of corresponding arma modeling is:

Wherein

the quadratic sum that means predicated error, the i.e. quadratic sum of the difference of the predicted value of each time series item and actual value.

3, model parameter estimation

In technical solution of the present invention, adopt the square method of estimation to be estimated the model parameter of ARMA (p, q).

For arbitrary finite sample data sequence x ₁, x ₂..., x _t, its sample autocovariance is defined as

${\widehat{\mathrm{\γ}}}_k=\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k},k=\mathrm{0,1,2},...,n-1$

Wherein ${\widehat{\mathrm{\γ}}}_0=\frac{1}{n}\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2$

Sample autocorrelation function is:

${\widehat{\mathrm{\ρ}}}_k=\frac{{\widehat{\mathrm{\γ}}}_k}{{\widehat{\mathrm{\γ}}}_0}=\frac{\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\frac{1}{n}\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2}=\frac{\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2},k=\mathrm{0,1,2},...,n-1$

The square of AR part is estimated as

Order

Covariance function is

With estimation replace γ _k, have

Can obtain parameter

To MA (q) model coefficient θ ₁, θ ₂..., θ _qthe employing square is estimated, is had

${\mathrm{\γ}}_0=(1+{\mathrm{\θ}}_1^2+{\mathrm{\θ}}_2^2+...+{\mathrm{\θ}}_q^2){\mathrm{\σ}}_a^2$

${\mathrm{\γ}}_k=(-{\mathrm{\θ}}_k+{\mathrm{\θ}}_1{\mathrm{\θ}}_{k+1}+...+{\mathrm{\θ}}_{q-k}{\mathrm{\θ}}_q){\mathrm{\σ}}_a^2,$

$k=\mathrm{1,2},...,m$

Below comprise altogether m+1 equation, for its parameter, equation is non-linear, adopts process of iteration to be solved.

Concrete steps are as follows, and equation is deformed into:

${\mathrm{\σ}}_a^2={\mathrm{\γ}}_0/(1+{\mathrm{\θ}}_1^2+{\mathrm{\θ}}_2^2+...+{\mathrm{\θ}}_q^2)$

${\mathrm{\θ}}_k=-\frac{{\mathrm{\γ}}_k}{{\mathrm{\σ}}_a^2}+{\mathrm{\θ}}_1{\mathrm{\θ}}_{k+1}+...+{\mathrm{\θ}}_{q-k}{\mathrm{\θ}}_q,$

$k=\mathrm{1,2}...,m$

Given θ ₁, θ ₂..., θ _qwith

one group of initial value, as θ ₁=θ ₂=...=θ _q=0,

the above two formula the right of substitution, the resulting value in the left side is first step iterative value, is designated as again this is worth to the right side of two formulas in substitution successively, just obtains the second step iterative value, the like, until adjacent twice iteration result while being less than given threshold value, got the approximate solution of the result of gained as parameter.

Solve the exponent number of time series models, will obtain the seasonal effect in time series predicted value; Obtain the seasonal effect in time series predicted value, must first set up concrete anticipation function; Set up concrete anticipation function, must know the exponent number of model.Seem so just to be absorbed in an endless loop, so how to address the above problem.

Rule of thumb, the time series models exponent number generally is no more than 5 rank.So when this algorithm specific implementation, at first hypothesized model is 1 rank, utilize method for parameter estimation in step 3 to obtain the parameter of first order modeling, and then set up estimation function and just can estimate to obtain each predicted value in the hope of the first order modeling time series models, thereby try to achieve the residual error variance of first order modeling; Afterwards, hypothesized model is second order, tries to achieve the residual error of second-order model with said method; By that analogy, can obtain the residual error of 1 to 5 rank model, select the exponent number of model of residual error minimum as the exponent number of final mask.

After the exponent number of model is determined, by step 3, can obtain the time series models for the prediction of photovoltaic generation power ultra-short term.

Wherein,

and θ _j(1≤j≤q) is coefficient, α _tit is the white noise sequence met the demands.

Finally it should be noted that: the foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, although with reference to previous embodiment, the present invention is had been described in detail, for a person skilled in the art, its technical scheme that still can put down in writing aforementioned each embodiment is modified, or part technical characterictic wherein is equal to replacement.Within the spirit and principles in the present invention all, any modification of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Claims (3)

1. the photovoltaic generation power ultra-short term Forecasting Methodology of a time-based series model, is characterized in that, comprises the following steps:

Step 1: gather the historical power data of photovoltaic plant, and historical power data is carried out to normalized, the normalized formula is as follows:

$x_i^{\′}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}$

Wherein, x _maxthe maximal value in historical power data, x _minthe minimum value in historical power data, x _ii historical power data, x ' _ii historical power data after normalization;

Step 2: set up fit equation according to historical power data after above-mentioned normalization, according to fit equation and the residual error variance set up, determine that the exponent number of model is the value of p and q;

Step 3: power data as historical as the photovoltaic plant gathered in step 1 utilizes data sequence x ₁, x ₂..., x _tmean, its sample autocovariance is defined as

${\widehat{\mathrm{\γ}}}_k=\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k},k=\mathrm{0,1,2},...,n-1,$

Wherein

x _tand x _t-kbe data sequence x ₁, x ₂..., x _tin numerical value;

Historical power data sample autocorrelation function is:

${\widehat{\mathrm{\ρ}}}_k=\frac{{\widehat{\mathrm{\γ}}}_k}{{\widehat{\mathrm{\γ}}}_0}=\frac{\frac{1}{n}\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\frac{1}{n}\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2}=\frac{\underset{t=k+1}{\overset{n}{\mathrm{\Σ}}}{x}_t{x}_{t-k}}{\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{x}_t^2},$

$k=\mathrm{0,1,2},...,n-1$

The square of autoregression part is estimated as

Order

Covariance function is

With

estimation replace γ _k,

Draw,

To moving average model coefficient θ ₁, θ ₂..., θ _qthe employing square is estimated, is had

${\mathrm{\γ}}_0=(1+{\mathrm{\θ}}_1^2+{\mathrm{\θ}}_2^2+...+{\mathrm{\θ}}_q^2){\mathrm{\σ}}_a^2$

${\mathrm{\γ}}_k=(-{\mathrm{\θ}}_k+{\mathrm{\θ}}_1+{\mathrm{\θ}}_{k+1}+...+{\mathrm{\θ}}_{q-k}{\mathrm{\θ}}_q){\mathrm{\σ}}_a^2,$

$k=\mathrm{1,2},...,m$

Below comprise altogether m+1 equation, for its parameter, equation is non-linear, adopts process of iteration to be solved and draws θ ₁, θ ₂..., θ _qvalue;

Step 4: the p drawn according to above-mentioned steps two and step 3 and q value and

, θ ₁, θ ₂..., θ _qvalue set up autoregressive moving-average model;

The formula of described autoregressive moving-average model is as follows:

Wherein,

and θ _j(1≤j≤q) is coefficient, α _tit is white noise sequence.

2. the photovoltaic generation power ultra-short term Forecasting Methodology of time-based series model according to claim 1, it is characterized in that, set up fit equation in above-mentioned steps two, the exponent number of determining model according to fit equation and the residual error variance of foundation is: the model increased progressively gradually with serial exponent number carrys out the matching original series, all calculates residual sum of squares (RSS) at every turn

then draw exponent number and

figure, when exponent number during by little increase,

can significantly descend, after reaching true exponent number

value tend towards stability gradually, the estimator of residual error variance is:

The observed value item number of " actual observed value number " actual use while referring to model of fit, for the sequence with N observed value, the matching autoregressive model, the actual observed value of using mostly is N-p most.

3. the photovoltaic generation power ultra-short term Forecasting Methodology of time-based series model according to claim 2, is characterized in that, above-mentioned matching autoregressive model formula is as follows:

Coefficient wherein

be called autoregressive coefficient, α _tmean residual sequence, and meet E (α _t)=0, α _tfor white noise sequence, { X _t, t=... ,-2 ,-1,0,1,2 ... mean the observed reading of random series, X _t-1, X _t-2..., X _t-pfor front p observed reading constantly.

Images