CN101236636A - Electric power equal price prediction method based on wavelet decomposition and phase space reconfiguration - Google Patents

Abstract

The invention discloses an electric power average price forecasting method based on wavelet decomposition and phase space reconstruction, which comprises the following steps that: history data of average price contribution rate of each fine-divided market is calculated according to the electric power sales data; the history data of the average price contribution rate is decomposed into a trend item, a period item and a random item by adopting the wavelet decomposition; an ARMA model, a period map method and a phase space reconstruction method are adopted to respectively forecast the trend item, the period item and the random item, the results are accumulated to obtain a forecasting result of the average price contribution rate; electric power average price of each fine-divided market is estimated according to the catalog electricity price; electric power market forecasting average price is calculated according to the average price estimation result and the forecasting result of the average price contribution rate of each fine-divided market. The electric power average price forecasting method effectively reduces the calculation amount of average price forecasting and improves the accuracy of electric power average price forecasting.

Description

Electric power equal price prediction method based on wavelet decomposition and phase space reconfiguration

Technical field

The present invention relates to a kind of Forecasting Methodology of electricity market electric power equal price, particularly a kind of electric power equal price prediction method based on wavelet decomposition and phase space reconfiguration.

Background technology

Power sale market can be divided into seven and segment market: big industry, non-general industry, resident's illumination, non-resident illumination, commerce, agricultural production, bulk sale.Average price is a COMPREHENSIVE CALCULATING amount of each electricity price that segments market, its definition be in certain unit interval all kinds of different load nature of electricity consumed electricity sales amount sales revenue sums of (as year, moon) whole society divided by whole society's sale electric weight sum the merchant.Computing formula is as follows:

$\stackrel{\‾}{P}=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{p}_i.q_i}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{q}_i},i=(\mathrm{1,2}...,n-1,n)$

Wherein

Be power sale market average price, p _iBe meant a certain electricity price in the listed power price system, q _iThen be meant the pairing of that month electricity sales amount of this electricity price.

Because average price has embodied the overall formation of whole society's power structure in fact, so average price prediction accurately will provide important evidence for the variation of analyzing power structure, and the result that can draw according to analysis, formulate and revise at each price strategy that segments market and service strategy, for the marketization of electric power enterprise provides strong intellectual support.And electric power enterprise is the control line loss, can the line loss index be proposed to the power department of subordinate, department of subordinate is not when touching the mark, may charge to total electricity sales amount to the line loss electric weight, cause the reduction of average price, average price prediction accurately will provide the data foundation for the management of line loss index, to reducing line loss Practical significance be arranged.

Because Chinese listed power price a multitude of names, reach 600 surplus kind.If according to traditional thinking, use average price computing formula prediction average price.Then need predict the pairing electricity sales amount of corresponding each item of listed power price, operand is very big first, and second the pairing electricity sales amount variation of some item is irregular, and precision of prediction is not high, and prediction effect gets half the result with twice the effort.

Summary of the invention

In order to solve the technical matters that existing electric power equal price prediction exists, the invention provides a kind of electric power equal price prediction method based on wavelet decomposition and phase space reconfiguration, the present invention is by introducing this intermediate quantity of average price contribution rate, prediction is converted into the average price prediction that respectively segments market, reduce calculated amount, improved precision of prediction.

The technical scheme that the present invention solves the problems of the technologies described above may further comprise the steps:

1), according to the power sale data computation average price contribution rate historical data that respectively segments market;

2), adopt wavelet decomposition that average price contribution rate historical data is decomposed into trend term, periodic term and random entry;

3), adopt arma modeling, period map method and phase space reconfiguration method anticipation trend item, periodic term and random entry respectively, with results added, obtain predicting the outcome of average price contribution rate;

4), estimate respectively to segment market electric power equal price according to listed power price;

5), predict the outcome calculating electricity market prediction average price according to respectively segment market average price estimated result and average price contribution rate.

Technique effect of the present invention is: the present invention has introduced the average price contribution rate, has reduced the calculated amount of prediction, in actual prediction, has obtained the better prediction precision.

Embodiment

The present invention has introduced the average price contribution rate, the average price prediction that forecasting problem is converted into the average price contribution rate and respectively segments market.The average price contribution rate is adopted wavelet decomposition, be decomposed into trend term, periodic term and random entry, according to above three kinds of seasonal effect in time series characteristics, adopt corresponding Forecasting Methodology to predict then.To the average price that respectively segments market, estimate according to listed power price.Its theoretical foundation is:

(1) definition of average price contribution rate

The average price contribution rate is used to show the weighing factor that respectively segments market to average price.

Find that in the excavation to data each segments market measurement for convenience to the weighing factor of average price, introduce an intermediate parameters-average price contribution rate α, α is defined as:

${\mathrm{\α}}_i=\frac{F_i/Q}{F/Q}\×100\%=\frac{F_i}{F}\×100\%$

I is the numbering that segments market in the formula, and F is the electricity charge, and Q is total electricity sales amount.

As can be seen from the above equation, be exactly Zong α segmented market in fact in every month electricity charge proportion in the electricity charge.

(2) derivation of average price predictor formula

It is that P, electric weight are that Q, the electricity charge are that F, average price are that P, average price contribution rate are α that bottom establishes electricity price.By definition as can be known:

$F=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{Q}_i{P}_i---\left(1\right)$

I is the numbering that segments market in the formula, i=(1,2...6,7)

$\stackrel{\‾}{P}=\frac{F}{Q}---\left(2\right)$

${\mathrm{\α}}_i=\frac{F_i/Q}{F/Q}\×100\%=\frac{F_i}{F}\×100\%---\left(3\right)$

According to above (1), (2), (3) formula, can derive:

∵ $F_i=\stackrel{\‾}{P_i}\×Q_i---\left(4\right)$

$\⇒{\mathrm{\α}}_i\frac{\stackrel{\‾}{P_i}\×Q_i}{\stackrel{\‾}{P}\×Q}\×100\%---\left(5\right)$

$\⇒\frac{{\mathrm{\α}}_i}{\stackrel{\‾}{P_i}}=\frac{Q_i}{\stackrel{\‾}{P}\×Q}\×100\%---\left(6\right)$

$\⇒\underset{i=1}{\overset{7}{\mathrm{\Σ}}}\frac{{\mathrm{\α}}_i}{\stackrel{\‾}{P_i}}=\underset{i=1}{\overset{7}{\mathrm{\Σ}}}\frac{Q_i}{\stackrel{\‾}{P}\×Q}\×100\%\frac{\underset{i=1}{\overset{7}{\mathrm{\Σ}}}{Q}_i}{\stackrel{\‾}{P}\×Q}=\frac{1}{\stackrel{\‾}{P}}---\left(7\right)$

$\⇒\stackrel{\‾}{P}=\frac{1}{\underset{i=1}{\overset{7}{\mathrm{\Σ}}}\frac{{\mathrm{\α}}_i}{\stackrel{\‾}{P_i}}}---\left(8\right)$

By (8) formula,, can derive the average price predicted value in whole power sale market if can predict average price contribution rate and the average price that respectively segments market.Like this, prediction just changes into the average price that respectively segments market and the prediction of average price contribution rate to the average price in whole power sale market.Below will analyze the average price contribution rate and the average price forecasting problem that respectively segment market respectively.

(3) prediction of average price contribution rate

To any nonlinear function f (t), can carry out multiple dimensioned wavelet decomposition to it:

$f\left(x\right)=\underset{j,k=-\∞}{\stackrel{\∞}{\mathrm{\Σ}}}{C}_{j,k}{\mathrm{\ψ}}_{j,k}\left(x\right)$

C wherein _{J, k}Be the antithesis of function f about ψ

The integration wavelet transformation, promptly

$C_{j,k}=2^{i/2}{\∫}_{-\∞}^{\∞}f\left(t\right)\widetilde{\mathrm{\ψ}}(2^{j}t-k)\mathrm{dt}$

{ ψ wherein _{J, k}It is wavelet basis.The average price contribution rate is carried out wavelet decomposition: J (t)=T (t)+M (t)+R (t) (T (t), M (t), R (t) represent trend, cycle, random entry respectively)

A) trend term T (t) prediction

Autoregression-moving averaging model (auto-regressive moving average, ARMA) modeling and forecasting adopted in the trend term prediction.

B) periodic term M (t) prediction

Periodic term can be regarded as by a plurality of sine functions and stochastic error and be formed by stacking, so:

$M\left(t\right)=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}(a_i\cos 2\mathrm{\π}{f}_{i}t+b_i\sin 2\mathrm{\π}{f}_{i}t)+{\mathrm{\ϵ}}_t$

Because: E (ε _t)=0

Parameter A is arranged: $A=\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{[M\left(t\right)-\underset{i=1}{\overset{k}{\mathrm{\Σ}}}(a_i\cos 2\mathrm{\π}{f}_{i}t+b_i\sin 2\mathrm{\π}{f}_{i}t)]}^2$

Adopt least square method to minimize A, obtain a _i, b _iValuation be:

It is as follows to get predicted value:

C) prediction of random entry

The random entry that obtains after the wavelet decomposition, because E (R) ≠ 0, so not directly be considered as white noise, will be in prediction as important one.Calculate its lyapunov index λ＞0, show that its chaotic characteristic is comparatively obvious, so adopt the Chaotic time series forecasting method of phase space reconfiguration to handle.

To random entry: R (t)=[r ₁, r ₂R _n], select suitable embedding dimension m and time delay τ, phase space reconfiguration:

R _i＝[r _i，r _i+τ，r _i+2τ，……，r _i+(m-1)τ]i＝1，2，...，n-(m-1)τ

With the trajectory of phase space last point as central point, distance center point nearer some tracing points as reference point, bigger than the near point assignment according to the distance of each point and central point for each point is given different weights, less than the far point assignment.These points are fitted, predict the trend of next locus of points:

$R_{m+1}=\frac{\underset{j=1}{\overset{j=m_p}{\mathrm{\Σ}}}{R}_{m_j}{e}^{-\mathrm{\α}(d_{m_j}-d_{\min })}}{\underset{j=1}{\overset{j=m_p}{\mathrm{\Σ}}}{e}^{-\mathrm{\α}(d_{m_j}-d_{\min })}}$

R _M+1For forecasting institute gets the track that descends any in the phase space, R _MjFor in the phase space a bit, m _pThen be the summation of counting, d _MjAnd d _MinBe space length and the minor increment of each point to central point.As can be seen from the above equation, decentering point distance is near more, and proportion is big more in prediction.At last can be from R _M+1In isolate predicted value r _M+1

(4) respectively segment market the determining of average price

Because the electricity consumption characteristics difference that respectively segments market, pricing strategy is also different.According to pricing strategy, 7 can be segmented market is divided into 3 classes.The first kind is industrial class, comprises big industry and non-general industry; Second class is civilian electric class, comprises residential electricity consumption, non-resident electricity consumption and commercial electricity consumption; The 3rd class then comprises bulk sale electricity consumption and agricultural production electricity consumption, and this class is subjected to weather influence bigger.

A) industrial class average price determines

Industry class electricity consumption can be divided into five electric pressures such as 200KV, 110KV, 35KV, 1-10KV and other, and preceding four electric pressures have clear and definite price in listed power price, and " other " class proportion is less, and electricity price can be estimated.Obtain each grade electricity price, calculate the average price contribution rate of each electric pressure then, repeat the prediction steps of above average price contribution rate, bring the average price computing formula into, the average price predicted value that gets final product at last.

B) civilian electric average price determines

Civilian electricity can be divided into following and two electric pressures of 1-10KV of 1KV, and each electric pressure also has clearly price in listed power price.Use and last same procedure, can obtain its average price predicted value.

C) bulk sale electricity consumption and agricultural production electricity consumption average price determines

These two the average price fluctuations that segment market are bigger, and the prediction difficulty is bigger.But because in whole electricity market, their proportions are less, use average electricity price of prior year substitutes as actual value, and is very little to whole precision of prediction influence.

Concrete steps of the present invention are:

1, data form database according to statistics, require to comprise each year, respectively segment market, the electricity sales amount and the electricity charge of each electric pressure;

2, according to raw data, calculating respectively segments market, the average price contribution rate of each electric pressure;

3, the prediction respectively segment market, each electric pressure average price contribution rate;

4, according to each electric pressure average price contribution rate predicted value, calculate the average price that big industry, non-general industry, resident's illumination, non-resident illumination, commercial electricity consumption etc. segment market, and with the average electricity price of preceding year bulk sale and agricultural production electricity consumption as the average price substitution value;

5, average price that will respectively segment market and average price contribution rate predicted value are brought the average price computing formula into, obtain the predicted value of electricity market average price.

Claims (1)

1. A kind of electric power equal price prediction method based on wavelet decomposition and phase space reconfiguration may further comprise the steps:

   1), according to the power sale data computation average price contribution rate historical data that respectively segments market;

   2), adopt wavelet decomposition that average price contribution rate historical data is decomposed into trend term, periodic term and random entry;

   3), adopt arma modeling, period map method and phase space reconfiguration method anticipation trend item, periodic term and random entry respectively, with results added, obtain predicting the outcome of average price contribution rate;

   4), estimate respectively to segment market electric power equal price according to listed power price;

   5), predict the outcome calculating electricity market prediction average price according to respectively segment market average price estimated result and average price contribution rate.