CN103136598B - Monthly calculation of power load machine Forecasting Methodology based on wavelet analysis - Google Patents

Abstract

The invention discloses a kind of monthly Methods of electric load forecasting based on wavelet analysis, including input module, wavelet transformation module, first kind prediction module, Equations of The Second Kind prediction module, wavelet reconstruction module, output module.Input module receives historical load datac ₀, and by historical datac ₀Send wavelet transformation module to；Historical data is decomposed by wavelet transformation module, resolves into successivelyd ₁、d ₂、d ₃、d ₄、c ₄Five components, whereind ₁‑d ₄There is the feature of annual cycles change,c ₄There is ever-increasing variation tendency；First kind prediction module paird ₁‑d ₄These four components are predicted；Equations of The Second Kind prediction module pairc ₄Component is predicted；Wavelet reconstruction module carries out wavelet reconstruction to each component after prediction, obtains predicting loadc ₀'；Finally by output module, prediction load is presented.

Description

Monthly calculation of power load machine Forecasting Methodology based on wavelet analysis

Technical field

The present invention relates to a kind of Methods of electric load forecasting, a kind of monthly calculation of power load machine based on wavelet analysis is pre- Survey method.

Background technology

Load prediction is power system development planning, fuel planning and the basis formulating generation schedule.Monthly prediction with the moon is Prediction period, using monthly power consumption as predictive content.By moon load prediction technology, it is possible to arranged rational power system Operational plan in mid-term, reduces operating cost, improves power supply reliability.

Month loading Changing Pattern in each year has annual periodicity, when carrying out moon load prediction, and must be in view of moon loading On the basis of annual periodicity, take full advantage of up-to-date monthly data, obvious to following predicted impact with up-to-date Changing Pattern Embody.Due to monthly the power consumption periodicity of Changing Pattern and short-term load forecasting and Day Load Curve Forecasting in each year Feature similar, therefore prior art uses the method being similar to short-term load forecasting mostly, including setting up autoregression-dynamic average (ARMA) model and artificial neural network method (ANN) etc..

Although said method has been obtained for being widely applied in the technical field of monthly load prediction, and achieve the most pre- Survey effect.But above-mentioned several method is only sensitiveer to variation characteristic in a short time, to long-term variation characteristic, such as exponential type Growth feature, can not reflect well.

Additionally, actual monthly load curve in addition to having with the fluctuation characteristic of seasonal variations, along with socioeconomic development and people People's growth in the living standard, also has the variation characteristic increased year by year.Therefore, regression model and the side of artificial neural network are used Method can reflect load curve seasonal fluctuations feature, but can not preferably reflect load curve growth trend year by year.

In order to reflect two kinds of variation characteristics of monthly load curve simultaneously, need to use certain method to be carried out by moon load signal curve point Solving, then the component to different characteristic is respectively adopted different Forecasting Methodologies and is predicted.

Summary of the invention

It is an object of the invention to according to providing one can change spy for a long time, year by year by effecting reaction in place of the deficiencies in the prior art Levy, more comprehensively monthly calculation of power load machine Forecasting Methodology based on wavelet analysis.

It is an object of the invention to be realized by following approach:

Monthly calculation of power load machine Forecasting Methodology based on wavelet analysis, it is characterized by, and comprises the steps:

Thering is provided load forecast module, it includes that input module, wavelet transformation module, first kind prediction module, Equations of The Second Kind are pre- Survey module, wavelet reconstruction module and output module；

Input module receives historical load data c from power system₀, and by historical data c₀Send wavelet transformation module to, Historical data is decomposed by wavelet transformation module, c₀Resolve into high fdrequency component d₁With low frequency component c₁, then by c₁Divide further Solution becomes d₂And c₂, c₂Further decompose into d₃And c₃, c₃Further decompose into d₄And c₄, thus obtain the d of decomposition₁、d₂、d₃、 d₄、c₄Five components, wherein d₁-d₄There is the feature of annual cycles change, c₄There is ever-increasing variation tendency；Use Need during wavelet transformation to carry out down-sampled to curve, often carry out a wavelet transformation, the sampled point half to be reduced of abscissa；

By d₁、d₂、d₃、d₄Component data send into first kind prediction module, first kind prediction module use can reflect data The data method of cyclic fluctuation feature to d₁、d₂、d₃、d₄These four components are predicted, and obtain d₁'、d₂'、d₃'、d₄' Four transform components；By c₄Data component sends into Equations of The Second Kind prediction module, and Equations of The Second Kind prediction module can reflect data sustainable growth The method of feature to c₄Component is predicted, and obtains transform component c₄', need when using wavelet reconstruction curve carries out a liter sampling, Often carry out a wavelet reconstruction, the sampled point half to be increased of abscissa.

Above-mentioned transform component is sent to wavelet reconstruction module, uses Mallat restructing algorithm, first by component d₄' and c₄' carry out weight Structure obtains c₃', the most by that analogy, by d₃' and c₃' further reconstruct obtain c₂', d₂' and c₂' further reconstruct obtain c₁', d₁' And c₁' further reconstruct obtain predicting load c₀'；

The prediction load c that will obtain₀' store, and exported by output module.

Wavelet transformation theory is proposed by Jean Morlet, has a wide range of applications in signal analysis field.Wavelet transformation can be by former Signal decomposition becomes the component of different scale such that it is able to reflect signal feature in different frequency domains and variation tendency.Compared to Fu In leaf transformation, wavelet transformation is better able to adapt to the signal that change is fierce, higher to the descriptive power of signal.

The continuous wavelet transform of signal f (t) can be expressed as:

$W(a,b)=\frac{1}{\sqrt{a}}\∫f\left(t\right)\mathrm{\ψ}\left(\frac{t-b}{a}\right)\mathrm{dt}$

Wherein a is scale parameter, and b is displacement parameter.

By the component of different scale is reconstructed, reducible signal, the formula of wavelet inverse transformation can be expressed as:

$f\left(t\right)=\frac{1}{C_{\mathrm{\ψ}}}{\∫}_0^{+\∞}{\∫}_{-\∞}^{+\∞}\frac{W(a,b)}{\sqrt{a}{a}^2}\mathrm{\ψ}\left(\frac{t-b}{a}\right)\mathrm{dadb}$

Wherein C_ψFor permissive condition.

Mallat et al. uses the method for wave filter to achieve wavelet transform, and concrete calculating process is:

$c_{j+1}\left(m\right)=\underset{m}{\mathrm{\Σ}}h(m-2k)c_j\left(m\right)$

$d_{j+1}\left(m\right)=\underset{m}{\mathrm{\Σ}}g(m-2k)c_j\left(m\right)$

Wherein j is decomposition scale, and k, m are translation coefficient, c_j+1(m) and d_j+1M () is the HFS of wavelet coefficient and low respectively Frequently part, h (m-2k) and g (m-2k) are high pass filter and low pass filter respectively.By using discrete wavelet analysis, permissible By original signal c₀Resolve into HFS d₁With low frequency part c₁, the most again by c₁Make to decompose further, resolve into d₂And c₂, By that analogy.By signal decomposition being become the component of different frequency, the feature of signal more details can be reflected, such that it is able to right Signal is preferably analyzed.

Owing to needing to carry out down-sampled to curve when using wavelet transformation, the yardstick after decomposing the most each time is all different , the therefore sampled point of abscissa half to be reduced.It was found that, after have passed through four layers of wavelet decomposition, low frequency component c₄Substantially, aperiodicity fluctuation tendency.Four layers of wavelet transformation by the cyclical component of load curve and persistently can increase as can be seen here Long component is separated: wavelet coefficient d₁-d₄With scale coefficient c₄Represent the basic variation tendency of load curve, wherein component d₁-d₄Having a strongest undulatory property, corresponding each year, load was with the variation tendency in season, component c₄Fluctuate less, and there is substantially increasing Long trend, corresponding is the characteristic of year load growth.Therefore can be well by the two of monthly load curve kinds by wavelet transformation Variation characteristic (with the fluctuation characteristic of seasonal variations and the variation characteristic that increases year by year) reflects, to monthly power load simultaneously Lotus has carried out comprehensive prediction, for formulating operation of power networks plan in mid-term, reduces operating cost, improves power supply reliability and provides number According to basis.

The present invention can further particularly as follows:

The data method of the described cyclic fluctuation feature that can reflect data or for time series method or for neural network.

First kind prediction module is primarily directed to four strong components of undulatory property and carries out processing calculating, it is therefore desirable to use the week of data The data processing method of phase property fluctuation characteristic realizes.Described time series method can be AR model, MA model, ARMA mould Type etc..Said method is not exhaustive, and the data processing method of all cyclic fluctuation features that can reflect data all can use.

The described method that can reflect data sustainable growth feature is trend extrapolation.

Equations of The Second Kind prediction module is primarily directed to c₄This stable cycle data of component carries out processing calculating, it is therefore desirable to use energy Enough reflecting the data processing method of data sustainable growth feature, described trend extrapolation can be the method for unbiased gray prediction, Gray prediction method is the one of trend extrapolation, the long term variations of its energy fitting data, and is predicted data.

The method have the advantages that

1. this Forecasting Methodology method by wavelet analysis, resolves into the component of different Changing Pattern by former load data.For not Same component can use different Forecasting Methodologies to be predicted, and can more accurately carry out a moon load prediction.

2. this Forecasting Methodology can reflect the cyclically-varying feature of monthly load data, and this is mainly by power load throughout the year Cyclically-varying cause.

3. this Forecasting Methodology can reflect the variation characteristic of monthly load data sustainable growth, and this is mainly due to socioeconomic Exhibition and the raising of living standards of the people, and cause power load to increase year by year.

Accompanying drawing explanation

Fig. 1 is the overall program flow diagram of monthly calculation of power load machine Forecasting Methodology based on wavelet analysis of the present invention.

Fig. 2 is the curve chart of each variable in Fig. 1.

Fig. 3 is wavelet transformation flow chart of the present invention.

Fig. 4 is wavelet reconstruction flow chart of the present invention.

Fig. 5 is RBF neural structure chart of the present invention.

Detailed description of the invention

See figures.1.and.2, based on wavelet analysis the monthly Methods of electric load forecasting of the present invention is including input module 1, little Wave conversion module 2, first kind prediction module 3,4,5,6, Equations of The Second Kind prediction module 7, wavelet reconstruction module 8, output module 9.Input module 1 receives historical load data c₀, and by historical data c₀Send wavelet transformation module 2 to；Wavelet transformation module 2 pairs of historical datas are decomposed, and resolve into d successively₁、d₂、d₃、d₄、c₄Five components, wherein d₁-d₄There is annual cycles Property change feature, c₄There is ever-increasing variation tendency；3,4,5,6 couples of d of first kind prediction module₁-d₄These four points Amount is predicted；Equations of The Second Kind prediction module 7 is to c₄Component is predicted；Each component after prediction is carried out by wavelet reconstruction module 8 Wavelet reconstruction, obtains predicting load c₀'；Finally by output module 9, prediction load is presented.

As it is shown on figure 3, in the present invention, wavelet transformation module 2 uses Mallat decomposition algorithm, first by former moon load data c₀ Resolve into high fdrequency component d₁With low frequency component c₁, the most by that analogy, by c₁Further decompose into d₂And c₂, c₂Divide further Solution becomes d₃And c₃, c₃Further decompose into d₄And c₄.Need to carry out down-sampled to curve when using wavelet transformation, divide each time Yardstick after solution is all different, often carries out a wavelet transformation, the sampled point half to be reduced of abscissa.

3,4,5,6 couples of d of first kind prediction module₁-d₄These four components are predicted, and obtain d₁'-d₄', selected prediction side Method can reflect the cyclic fluctuation feature of data, specifically includes time series method (AR model, MA model, arma modeling Deng), neural network etc..The present embodiment is to coefficient d₁-d₄When being predicted, use RBF neural prediction side Method.

RBF neural is a kind of three layers of feedforward network, as shown in Figure 5.It comprises input layer X={x₁,x₂...x_n, hidden layer With output layer y, hidden layer is made up of one group of RBF, and hidden layer node comprises two important parameters: center C_iAnd width σ_i, RBF generally chooses Gaussian function.The training of RBF neural realizes each base letter by minimizing object function Weights ω between number and output node_iRegulation.Training method includes clustering procedure, method of least square, gradient method etc., adopts here Use clustering procedure.When using above-mentioned RBF neural to be predicted, input vector X is historical data, and output vector Y is Single predictive value, in order to be predicted multiple points simultaneously, then needs to repeat to set up the most above-mentioned RBF neural.

Use RBF neural to d₁-d₄It is predicted, result d as shown in Figure 2₁'-d₄' curve chart.

Equations of The Second Kind prediction module 7 is to c₄Component is predicted, and obtains c₄', selected Forecasting Methodology can reflect that data persistently increase Long feature, including all kinds of trend extrapolations, as dynamic average, exponential smoothing, gray prediction etc..

Gray prediction method is the one of trend extrapolation, the long term variations of its energy fitting data, and is predicted data.

To coefficient c₄When being predicted, owing to unbiased GM (1,1) model is more accurate compared to Traditional GM (1,1) model, this Literary composition uses the method for unbiased gray prediction.

Assume that sequence to be predicted is x⁽⁰⁾={x⁽⁰⁾(1),x⁽⁰⁾(2),...,x⁽⁰⁾(n)}.First to x⁽⁰⁾Carry out cumulative summation formation sequence x⁽¹⁾={x⁽¹⁾(1),x⁽¹⁾(2),...,x⁽¹⁾(n) }, wherein:

$x^{\left(1\right)}\left(i\right)=\underset{j=1}{\overset{i}{\mathrm{\Σ}}}{x}^{\left(0\right)}\left(j\right),i=\mathrm{1,2,3}...$

According to gray system theory, cumulative sequence x⁽¹⁾There is exponential increase rule, it is believed that x⁽¹⁾Meet following Differential Equation Model:

$\frac{{\mathrm{dx}}^{\left(1\right)}}{\mathrm{dt}}+{\mathrm{ax}}^{\left(1\right)}=b$

If above formula uses discrete form to represent, can turn to:

$x^{\left(0\right)}(k+1)+\frac{1}{2}a[x^{\left(1\right)}\left(k\right)+x^{\left(1\right)}(k+1)]=b,k=\mathrm{1,2,3}...$

Write as matrix form to have:

$\left[\begin{array}{c}{x}^{\left(0\right)}\left(2\right)\\ x^{\left(0\right)}\left(3\right)\\ \·\\ \·\\ \·\\ x^{\left(0\right)}\left(n\right)\end{array}\right]=\left[\begin{array}{cc}-\frac{1}{2}[x^{\left(1\right)}\left(1\right)+x^{\left(1\right)}\left(2\right)]& 1\\ -\frac{1}{2}[x^{\left(1\right)}\left(2\right)+x^{\left(1\right)}\left(3\right)]& 1\\ \·& \·\\ \·& \·\\ \·& \·\\ -\frac{1}{2}[x^{\left(1\right)}(n-1)+x^{\left(1\right)}\left(n\right)]& 1\end{array}\right]\left[\begin{array}{c}a\\ b\end{array}\right];$ It is abbreviated as: $Y_n=B\·\left[\begin{array}{c}a\\ b\end{array}\right]$

Wherein:

$Y_n=\left[\begin{array}{c}{x}^{\left(0\right)}\left(2\right)\\ x^{\left(0\right)}\left(3\right)\\ \·\\ \·\\ \·\\ x^{\left(0\right)}\left(n\right)\end{array}\right],$ $B=\left[\begin{array}{cc}-\frac{1}{2}[x^{\left(1\right)}\left(1\right)+x^{\left(1\right)}\left(2\right)]& 1\\ -\frac{1}{2}[x^{\left(1\right)}\left(2\right)+x^{\left(1\right)}\left(3\right)]& 1\\ \·& \·\\ \·& \·\\ \·& \·\\ -\frac{1}{2}[x^{\left(1\right)}(n-1)+x^{\left(1\right)}\left(n\right)]& 1\end{array}\right]$

Above-mentioned equation utilizes method of least square can obtain the least square solution of matrix A:

$\left[\begin{array}{c}\hat{a}\\ \hat{b}\end{array}\right]={\left(B^{T}B\right)}^{-1}{B}^T{Y}_n$

According to the modeling method of unbiased GM (1,1) model, need byWithMake calculated as below:

${\hat{a}}^{\′}=\ln \frac{2-\hat{a}}{2+\hat{a}},$ $A=\frac{2\hat{b}}{2+\hat{a}}$

Sequence to be measured can represent in order to drag:

$x^{\left(0\right)}(k+1)={\mathrm{Ae}}^{{\hat{a}}^{\′}\·k}$

Use unbiased GM (1,1) model to coefficient c₄C in the result being predicted such as Fig. 2₄' curve chart.

As shown in Figure 4, in the present invention, wavelet reconstruction module 8 uses Mallat restructing algorithm, first by component d₄' and c₄' carry out weight Structure obtains c₃', the most by that analogy, by d₃' and c₃' further reconstruct obtain c₂', d₂' and c₂' further reconstruct obtain c₁', d₁' And c₁' further reconstruct obtain predicting load c₀'.Need when using wavelet reconstruction curve carries out a liter sampling, often carry out the least Reconstructed wave, the sampled point half to be increased of abscissa.

Input module 1 and input module 9 are realized by software.By input module 1, operator can import a moon load and go through History data c₀, and be stored in data base.Output module 9 can derive prediction load c₀', and and then draw prediction load curve, Thus conveniently observe.

The not described part of the present invention is same as the prior art.

Claims (3)

1. monthly calculation of power load machine Forecasting Methodology based on wavelet analysis, it is characterised in that comprise the steps:

Thering is provided load forecast module, it includes input module, wavelet transformation module, first kind prediction module, Equations of The Second Kind prediction module, wavelet reconstruction module and output module；

Input module receives the historical load data from power systemc ₀, and by historical datac ₀Sending wavelet transformation module to, historical data is decomposed by wavelet transformation module,c ₀Resolve into high fdrequency componentd ₁And low frequency componentc ₁, then willc ₁Further decompose intod ₂Withc ₂,c ₂Further decompose intod ₃Withc ₃,c ₃Further decompose intod ₄Withc ₄, thus obtain decompositiond ₁、d ₂、d ₃、d ₄、c ₄Five components, whereind ₁-d ₄There is the feature of annual cycles change,c ₄There is ever-increasing variation tendency；Need to carry out down-sampled to curve when using wavelet transformation, often carry out a wavelet transformation, the sampled point half to be reduced of abscissa；

By component datad ₁、d ₂、d ₃、d ₄Sending into first kind prediction module, first kind prediction module uses the data method pair of the cyclic fluctuation feature that can reflect datad ₁、d ₂、d ₃、d ₄These four components are predicted, and obtaind ₁'、d ₂'、d ₃'、d ₄' four transform component；Willc ₄Data component sends into Equations of The Second Kind prediction module, and Equations of The Second Kind prediction module can reflect the method pair of the feature of data sustainable growthc ₄Component is predicted, and obtains transform componentc ₄', need when using wavelet reconstruction curve carries out a liter sampling, often carry out a wavelet reconstruction, the sampled point half to be increased of abscissa；

Above-mentioned transform component is sent to wavelet reconstruction module, uses Mallat restructing algorithm, first by componentd ₄' andc ₄' be reconstructed and obtainc ₃', the most by that analogy, willd ₃' andc ₃' further reconstruct obtainc ₂',d ₂' andc ₂' further reconstruct obtainc ₁',d ₁' andc ₁' further reconstruct obtain predicting loadc ₀'；

The prediction load that will obtainc ₀' store, and exported by output module.

Monthly calculation of power load machine Forecasting Methodology based on wavelet analysis the most according to claim 1, it is characterised in that the data method of the described cyclic fluctuation feature that can reflect data or for time series method or for neural network.

Monthly calculation of power load machine Forecasting Methodology based on wavelet analysis the most according to claim 1, it is characterised in that the described method that can reflect data sustainable growth feature is trend extrapolation.