CN103679287A - Combined type power load forecasting method - Google Patents

Abstract

The invention discloses a combined type power load forecasting method which comprises an equal-weight average combined forecasting method, a covariance combined forecasting method and a regression combined forecasting method, and after the three forecasting methods are combined, the combined type power load forecasting method is generated. Forecasting results are combined so that a combined forecasting result can be obtained. Information of multiple forecasting models is collected in the combined forecasting result so that the purpose of the improvement of the forecasting result can be achieved. Compared with the prior art, the combined type power load forecasting method that appropriate weights of the forecasting results obtained through the forecasting methods are selected so as to be weighed and averaged is adopted or the forecasting methods are compared, so that the forecasting method with the optimal goodness of fit or the smallest standard deviation is selected. Thus, the combined type power load forecasting method is simple in algorithm, capable of rapidly and efficiently improving the work efficiency and enabling the algorithm to be rapid and accurate and high in popularization and application value.

Description

A kind of combined Methods of electric load forecasting

Technical field

The present invention relates to a kind of Methods of electric load forecasting, relate in particular to a kind of combined Methods of electric load forecasting.

Background technology

Load forecast is exactly the condition of the key factor impacts such as taking into account system operation characteristic, natural conditions, social condition and regional economic development situation, utilize historical load value through a series of mathematical computations, in the situation that meeting certain precision, determine the load value of following certain particular moment.

Load forecast is the important component part of the various technical safety measuress of electric system, and it is the same with relay protection, stability Calculation, short circuit calculation, and the safety of electric system, economy, stable operation are had to very important effect.Coming into the market economic today, especially be faced with power market reform, load prediction accurately can arranged rational unit operation capacity, improve the utilization factor of unit, reduce necessary spinning reserve capacity, reduce energy loss, guarantee the quality of power supply, thereby effectively reduce cost of electricity-generating and improve the economic and social benefits.And the computing of Methods of electric load forecasting of the prior art is complicated and effect is not very desirable, inefficiency, is difficult to promote.

Summary of the invention

Object of the present invention provides a kind of combined Methods of electric load forecasting with regard to being in order to address the above problem.

The present invention is achieved through the following technical solutions above-mentioned purpose:

Weight average combined prediction method, covariance combined prediction method and the regression combination predicted method such as the present invention includes, the described weight average combined prediction method that waits is:

If f _i(i=1,2, k) is the predicted value of i model, if fc represents combined prediction value, the combined prediction that waits weight average combined prediction method to obtain is

$f_c=\frac{1}{k}\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{f}_i;$

Described covariance combined prediction method is:

If f1, f2 be two about f without inclined to one side predicted value, fc is average weighted combined prediction value, if predicated error is respectively e1, e2 and ec, getting w1, w2 is corresponding weight coefficient, and there is w1+w2=1, have fc=w1f1+w2f2, requiring fc is also without inclined to one side, and error and variance thereof are respectively ec=w1e1+w2e2;

Described regression combination predicted method is:

Be located at moment t about y _ttwo predicted values be f _t ⁽¹⁾, f _t ⁽²⁾, be constructed as follows the combination forecasting of recurrence form:

$\left.\begin{array}{c}{y}_t={{\mathrm{\β}}_1{f}_t}^{\left(1\right)}+{{\mathrm{\β}}_2{f}_t}^{\left(2\right)}+{\mathrm{\ϵ}}_{\mathrm{\τ}}\\ {\mathrm{\β}}_1+{\mathrm{\β}}_2=1\end{array}\right\}$

In formula: f _t ⁽¹⁾, f _t ⁽²⁾be two groups of predicted values, β ₁, β ₂for weight coefficient, ε _τfor error term, consider constraint condition β ₁+ β ₂=1, there is [y _t-f _t ⁽²⁾]=β ₁[f _t ⁽¹⁾-f _t ⁽²⁾]+ε _τapplication least square method, can obtain:

$\mathrm{\β}=\frac{\underset{t}{\mathrm{\Σ}}[{f_t}^{\left(1\right)}-{f_t}^{\left(2\right)}][y_t-{f_t}^{\left(2\right)}]}{\underset{t}{\mathrm{\Σ}}{[{f_t}^{\left(t\right)}-{f_t}^{\left(2\right)}]}^2}$

In conjunction with covariance combined prediction method, obtain w ₁=β ₁, w ₂=β ₂.

Beneficial effect of the present invention is:

The present invention is a kind of combined Methods of electric load forecasting, compared with prior art, the present invention's employing is chosen suitable weight by predicting the outcome of several Forecasting Methodology gained and is weighted average Forecasting Methodology or compares in several Forecasting Methodologies, select the Forecasting Methodology of goodness of fit the best or standard deviation minimum, therefore algorithm of the present invention simple, can increase work efficiency fast and effectively, and algorithm fast accurate, has the value of promoting the use of.

Embodiment

Below the invention will be further described:

Weight average combined prediction method, covariance combined prediction method and the regression combination predicted method such as the present invention includes, the described weight average combined prediction method that waits is:

If f _i(i=1,2, k) is the predicted value of i model, if fc represents combined prediction value, the combined prediction that waits weight average combined prediction method to obtain is

$f_c=\frac{1}{k}\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{f}_i;$

The precision of prediction that does not need to understand Individual forecast value fi Deng weight average combined prediction method, do not need to know the mutual relationship between the error of Individual forecast yet, method is simple, is in the situation that to the unknown completely of the precision of prediction of various Forecasting Methodologies, a kind of safer method of taking.

Described covariance combined prediction method is:

If f1, f2 be two about f without inclined to one side predicted value, fc is average weighted combined prediction value, if predicated error is respectively e1, e2 and ec, getting w1, w2 is corresponding weight coefficient, and there is w1+w2=1, have fc=w1f1+w2f2, requiring fc is also without inclined to one side, and error and variance thereof are respectively ec=w1e1+w2e2;

Described regression combination predicted method is:

Be located at moment t about y _ttwo predicted values be f _t ⁽¹⁾, f _t ⁽²⁾, be constructed as follows the combination forecasting of recurrence form:

$\left.\begin{array}{c}{y}_t={{\mathrm{\β}}_1{f}_t}^{\left(1\right)}+{{\mathrm{\β}}_2{f}_t}^{\left(2\right)}+{\mathrm{\ϵ}}_{\mathrm{\τ}}\\ {\mathrm{\β}}_1+{\mathrm{\β}}_2=1\end{array}\right\}$

In formula: f _t ⁽¹⁾, f _t ⁽²⁾be two groups of predicted values, β ₁, β ₂for weight coefficient, ε _τfor error term, consider constraint condition β ₁+ β ₂=1, there is [y _t-f _t ⁽²⁾]=β ₁[f _t ⁽¹⁾-f _t ⁽²⁾]+ε _τapplication least square method, can obtain:

$\mathrm{\β}=\frac{\underset{t}{\mathrm{\Σ}}[{f_t}^{\left(1\right)}-{f_t}^{\left(2\right)}][y_t-{f_t}^{\left(2\right)}]}{\underset{t}{\mathrm{\Σ}}{[{f_t}^{\left(t\right)}-{f_t}^{\left(2\right)}]}^2}$

In conjunction with covariance combined prediction method, obtain w ₁=β ₁, w ₂=β ₂.

Above three kinds of Forecasting Methodologies in conjunction with rear be combination forecasting method, single Forecasting Methodology information used is limited, and different Forecasting Methodologies information used is incomplete same, various Individual forecast results are combined and can obtain a kind of combined prediction result, the information of the multiple forecast model of combined prediction result set, thereby can reach the object that improvement predicts the outcome, but also should be noted, combined prediction be can not right-on description premeasuring at Individual forecast model Changing Pattern time play a role, find a model that reflects practical development rule completely to predict, can effectively predict electric load.

Claims (1)

1. a combined Methods of electric load forecasting, is characterized in that, weight average combined prediction method, covariance combined prediction method and the regression combination predicted method such as comprises, the described weight average combined prediction method that waits is:

If f _i(i=1,2, k) is the predicted value of i model, if fc represents combined prediction value, the combined prediction that waits weight average combined prediction method to obtain is

$f_c=\frac{1}{k}\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{f}_i;$

Described covariance combined prediction method is:

If f1, f2 be two about f without inclined to one side predicted value, fc is average weighted combined prediction value, if predicated error is respectively e1, e2 and ec, getting w1, w2 is corresponding weight coefficient, and there is w1+w2=1, have fc=w1f1+w2f2, requiring fc is also without inclined to one side, and error and variance thereof are respectively ec=w1e1+w2e2;

Described regression combination predicted method is:

Be located at moment t about y _ttwo predicted values be f _t ⁽¹⁾, f _t ⁽²⁾, be constructed as follows the combination forecasting of recurrence form:

$\left.\begin{array}{c}{y}_t={{\mathrm{\β}}_1{f}_t}^{\left(1\right)}+{{\mathrm{\β}}_2{f}_t}^{\left(2\right)}+{\mathrm{\ϵ}}_{\mathrm{\τ}}\\ {\mathrm{\β}}_1+{\mathrm{\β}}_2=1\end{array}\right\}$

In formula: f _t ⁽¹⁾, f _t ⁽²⁾be two groups of predicted values, β ₁, β ₂for weight coefficient, ε _τfor error term, consider constraint condition β ₁+ β ₂=1, there is [y _t-f _t ⁽²⁾]=β ₁[f _t ⁽¹⁾-f _t ⁽²⁾]+ε _τapplication least square method, can obtain:

$\mathrm{\β}=\frac{\underset{t}{\mathrm{\Σ}}[{f_t}^{\left(1\right)}-{f_t}^{\left(2\right)}][y_t-{f_t}^{\left(2\right)}]}{\underset{t}{\mathrm{\Σ}}{[{f_t}^{\left(t\right)}-{f_t}^{\left(2\right)}]}^2}$

In conjunction with covariance combined prediction method, obtain w ₁=β ₁, w ₂=β ₂.