CN103258069A - Forecasting method for power demand of iron and steel industry - Google Patents

Abstract

The invention discloses a forecasting method for power demand of the iron and steel industry and load forecasting of a power system. The forecasting method includes classifying influencing factors of electricity consumption of the iron and steel industry, analyzing key influencing factors of the electricity consumption of the iron and steel industry by a qualitative and quantitative combined method, proposing an index system of the electricity consumption of the iron and steel industry and providing a demand forecasting model. The forecasting method provides a scientific and practical technical scheme for power enterprises to accurately forecast the power demand of related industries and further forecast the power demand of the whole society. The forecasting method has the advantages that existing power demand forecasting methods are enriched and improved, and the forecasting method is significant in improving power demand forecasting level of the power enterprises and can generate remarkable economic and social benefits.

Description

A kind of Forecasting Methodology of steel industry electricity needs

Technical field

The present invention relates to the load prediction of electric system, relate in particular to the prediction of steel industry electricity needs.

Background technology

Electric power demand forecasting is the element task of departments such as planning in the electric system, plan, electricity consumption, scheduling, and in the process of power industry marketization operation, electric power demand forecasting becomes one of core business of departments such as marketing, the marketing again.

Iron and steel, non-ferrous metal, chemical industry, nonmetalliferous ore Tetramune four big highly energy-consuming trade power consumption amounts account for 1/3 of whole society's power consumption, and the highly energy-consuming trade power consumption has the characteristics of big rise and big fall, and the increase and decrease of its power consumption has material impact to the electricity consumption aggregate demand.The highly energy-consuming trade power consumption has following characteristics:

(1) it is big to account for the influence great, that electricity consumption is increased of the ratio of whole society's electricity consumption.Be example with 2010, the proportion that iron and steel, non-ferrous metal, chemical industry and nonmetalliferous ore Tetramune industry four big highly energy-consuming trade power consumptions account for whole society's electricity consumption in the State Grid Corporation of China operation zone is that 32.1%, four big highly energy-consuming industry is 34% to the contribution rate that whole society's electricity consumption increases.This shows that fully the highly energy-consuming electricity consumption is to the electricity consumption situation important influence in company management zone, and prediction highly energy-consuming trade power consumption is significant to whole society's electricity consumption prediction.

(2) be subjected to macroeconomic influence big, fluctuating range big, the prediction difficulty is big.The power consumption growth is subjected to macroeconomic very big influence, and power consumption changes with the economic situation degree of correlation higher.Before in May, 2008, the monthly speedup of highly energy-consuming trade power consumption generally is higher than whole society's electricity consumption speedup.After in May, 2008, be subjected to the influence of financial crisis, gliding all appears in whole society's electricity consumption and highly energy-consuming electricity consumption, but the influence that the highly energy-consuming trade power consumption is subjected to is bigger, and the range of decrease is also bigger.After economy bottomed out in 2009, the rebound momentum of highly energy-consuming trade power consumption was again faster than whole society's electricity consumption.The characteristics that influenced by economic situation and occur rising and fall sharply and quickly make the difficulty of prediction highly energy-consuming trade power consumption strengthen, and adopt the Forecasting Methodology of historical data extrapolation to be difficult to obtain desirable prediction effect especially.We can say that as long as can predict highly energy-consuming trade power consumption amount more exactly, electric power demand forecasting just hits half.

(3) variation tendency of highly energy-consuming trade power consumption is slightly led over whole society's electricity consumption.With company management zone power consumption data instance in 2008, highly energy-consuming trade power consumption amount reached peak value in May, begin to glide month by month after June.And whole society's electricity consumption and commercial power just reach peak value in July, begin subsequently to descend.The variation of highly energy-consuming trade power consumption has taken the lead 2 months than whole society electricity consumption.

Traditional electric power demand forecasting technology can be divided into two big classes: a class is the forecasting techniques according to electricity needs self historical data, as extrapolation method, time series method etc.Another kind of is the forecasting techniques of consideration and influence factor correlativity, as elastic coefficient method, output value unit consumption method, regression analysis etc.But these forecasting techniquess are not all considered the characteristics of highly energy-consuming trade power consumption, can't reflect the inherent law that the highly energy-consuming trade power consumption changes, and the analyses and prediction effect is also undesirable.

Domestic and foreign literature has some qualitative analyses to highly energy-consuming trade power consumption characteristic, but quantitative test is few, does not still have available analysis, forecast model.

Current, electric power enterprise is more and more recognized analysis, prediction highly energy-consuming trade power consumption to the significance of prediction whole society electricity consumption, but because the restriction of theoretical and method, also there is deficiency in electric power enterprise to analysis, the prediction of highly energy-consuming trade power consumption at present.

(1) understanding to highly energy-consuming trade power consumption influence factor remains deeply, does not form unified, analysis of science index system.The factor that influences the highly energy-consuming trade power consumption is very complicated, and some is direct factor, and some is indirect factor.Because the understanding to influence factor is not goed deep into, and causes present constituent parts that there is very big-difference in the analytic angle of highly energy-consuming trade power consumption, such as the analysis industry factor that has, the analysis macroeconomy factor that has, the analysing output that has, the analysis price again that has.Do not have the unified analysis indexes of a cover, cause the analysis to the highly energy-consuming trade power consumption to be difficult to deeply.

(2) do not grasp the quantitative relationship of highly energy-consuming trade power consumption and influence factor.Such as: well-known, the steel industry power consumption mainly depends on iron and steel output, and iron and steel output depends on the iron and steel demand of downstream industries such as real estate, automobile, but the quantitative relationship of iron and steel output and downstream industry demand how, but there is not available achievement in research, therefore present analysis or prediction to the highly energy-consuming industry is qualitatively mostly, is difficult to carry out quantitative test or prediction.

(3) lack Forecasting Methodology targetedly.Predict that exactly the highly energy-consuming trade power consumption has significance to improving whole society's electricity consumption prediction accuracy.Though the achievement of the relevant electric power demand forecasting method of domestic and foreign literature is a lot of at present, but these Forecasting Methodologies are not considered the characteristics of highly energy-consuming trade power consumption basically, specific aim and validity are not strong, and this causes the analyst to lack effective method and model when prediction highly energy-consuming trade power consumption.Sometimes using certain methods to obtain predicting the outcome reluctantly, also is " knowing that sth. is so but not why is so ", does not know the reason that the highly energy-consuming trade power consumption changes.

Summary of the invention

The Forecasting Methodology that the purpose of this invention is to provide a kind of steel industry electricity needs, this method is classified to the influence factor of steel industry electricity consumption, adopt the key influence factor of qualitative and the methods analyst steel industry electricity consumption that quantitatively combines, proposed to analyze the index system of steel industry electricity consumption, and provided Demand Forecast Model, accurately predict relevant industries electricity needs and then prediction whole society electricity needs for electric power enterprise the technical scheme of a kind of science, practicality is provided.

In order to achieve the above object, the present invention adopts following technical scheme:

Influence factor to the steel industry electricity consumption is classified, adopt qualitative and the methods analyst that quantitatively combines the key influence factor of steel industry electricity consumption, assessed the degree of correlation of influence factor and power consumption, proposed to analyze the index system of steel industry electricity consumption, be conducive to electric power enterprise and hold the inherent law that the steel industry electricity consumption increases.

Employing meets the Forecasting Methodology of highly energy-consuming industry characteristic, has set up the electric power demand forecasting model of steel industry, has checked the validity of model, has assessed the precision of prediction of model.Forecast model has very strong specific aim and practicality, and prediction effect is good.

Advantage of the present invention is: enriches and perfect existing electric power demand forecasting method, and significant to the electric power demand forecasting level that improves electric power enterprise, can produce remarkable economic efficiency and social benefit:

(1) is applied to power planning work, improves the accuracy of load prediction in the planning, be conducive to optimize electric power project investment scale, reduce the investment waste.

(2) be applied to plan, scheduling and the market analysis work of power grid enterprises, can improve the accuracy of market analysis, avoid occurring that supply falls short of demand and serious easy situation, reduce the loss that electric power supply is not enough and electric power overcapitalizes and brings.

(3) be applied to generation schedule and the coal procurement plan of electricity power enterprise, can help electricity power enterprise to optimize electricity and coal reserve, reduce the electric coal occupation of capital, avoid electric coal shortage to cause the generating capacity deficiency simultaneously.

(4) electric power enterprise can be developed the electric power demand forecasting software product according to the present invention.

Embodiment

Technology path of the present invention is as follows:

1. the basic step of electric power demand forecasting

Electric power demand forecasting is divided into following step:

(1) target of prediction and predictive content determines;

(2) collection of relevant historical data;

(3) analysis of basic data;

The prediction of (4) electric system correlative factor data or obtain;

(5) selection of forecast model and method and choice;

(6) modeling;

(7) data pre-service;

(8) identification of Model Parameters;

(9) model evaluation, the conspicuousness of testing model;

(10) application model is predicted;

(11) analysis-by-synthesis that predicts the outcome and evaluation.

2. Regression Forecast

Electricity needs is to be determined by economic development level, and the regression forecasting class model is just by setting up the correlationship between electricity needs and the economic variable, realizes seizure to the electricity needs rule of development with Regression Forecasting Technology.Regression Forecast is by the analysis and research to historical data, explore inner link and the development and change rule of economic, social each related factors and electricity needs, and according to in project period, following electricity needs is calculated in the prediction of this area economy, social development situation, and its task is the relation of determining between predicted value and the factor of influence.Regression Forecast is the development of principle of least square method, can be divided into one-variable linear regression, multiple linear regression and nonlinear regression model (NLRM) at present.

1. Linear Regression Model in One Unknown

The Linear Regression Model in One Unknown expression formula is as follows:

y＝f(S,X)＝a+bx+ε

The parameter vector of S in the formula---model, S=[a, b] ^T

X---independent variable;

Y---depend on the stochastic variable of x;

ε---Normal Distribution N (0, σ ²) stochastic error, be called random disturbance again.

Residual sum of squares (RSS) is:

$Q(a,b)=\stackrel{n}{\mathrm{\Σ}}{(y_i-a-{\mathrm{bx}}_i)}^2,(i=\mathrm{1,2},...,n)$

X in the formula _i, y _i---sample.

Utilize least square method to come estimation model parameter a, b, so that Q reaches minimal value, obtain the model parameter estimation value and be:

$\left\{\begin{array}{c}\hat{b}=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\\ \hat{a}=\stackrel{\‾}{y}-\hat{b}\stackrel{\‾}{x}\end{array}\right.$

Wherein $\stackrel{\‾}{x}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_i;$ $\stackrel{\‾}{y}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{y}_i,$ Predictive equation is:

$\hat{y}=\hat{a}+\hat{b}x$

2. multiple linear regression model

If y is explained variable, y _iThe i time observed reading y=(y for y ₁, y ₂... y _n) ^T, k explanatory variable x=(x ₁, x ₂... x _k), β=(β ₀, β ₁.. β _k) ^TBe regression coefficient, ε=(ε ₁, ε ₂.. ε _n) ^TThe expression stochastic error.

The observation matrix of note explanatory variable is $X=\left[\begin{array}{cccc}1& x_{11}& \·\·\·& x_{1k}\\ 1& x_{21}& \·\·\·& x_{2k}\\ \·& \·& \mathrm{}& \·\\ \·& \·& \mathrm{}& \·\\ \·& \·& \mathrm{}& \·\\ 1& x_{\mathrm{nl}}& \·\·\·& x_{3k}\end{array}\right],$ Then multiple linear regression model can be write as the form of matrix:

y＝Xβ+ε

(1) OLS estimates

1. the estimated value of regression coefficient β

$\widehat{\mathrm{\β}}={\left(X^{T}X\right)}^{-1}{X}^{T}Y$

Satisfy character: $E\left(\widehat{\mathrm{\β}}\right)=\mathrm{\β},\mathrm{cov}\left(\widehat{\mathrm{\β}}\right)={\mathrm{\σ}}^2{\left(X^{T}X\right)}^{-1}$

2. stochastic error ε

Each element is obeyed N (0, σ among the stochastic error ε ²) normal distribution.The estimated value of σ is :

${\widehat{\mathrm{\σ}}}^2=\frac{e^{T}e}{n-k-1}---(1-14)$

(2) linear dependence significance test

1. related coefficient (coefficient of determination)

According to variance analysis, $S_T^2=S_R^2+S_E^2,$

Related coefficient:

Related coefficient can be used for describing the linear dependence degree of y and x, and related coefficient is more near 1, and then fitting effect is more good.

2. F check

F check shows whether y and x linear dependence be remarkable.

Suppose H ₀: β ₁=β ₂=... β _k=0

If null hypothesis is set up, the definition statistic: $F=\frac{\frac{S_R^2}{k}}{\frac{S_E^2}{n-k-1}}~F(k,n-k-1)$ Obeying degree of freedom is that (k, F n-k-1) distributes, under certain confidence level a, if F＞F _aThe refusal null hypothesis, otherwise accept null hypothesis.

3. t check

The t check is whether each explanatory variable of check is remarkable to the influence degree of explained variable,

Suppose H ₀: β _i=0,

If null hypothesis is set up, the definition statistic: $t=\frac{{\widehat{\mathrm{\β}}}_i}{\sqrt{{\widehat{\mathrm{\σ}}}^2{\left(X^{T}X\right)}_{\mathrm{ii}}^{-1}}}~t(n-k-1),$ Under certain confidence level a, if t＞t _aThe refusal null hypothesis, otherwise accept null hypothesis.

(3) prediction

The value of the n+1 phase of given explanatory variable is: C=(1, x _{N+l, 1}.., x _{N+l, k}) ^TSo, y _N+lConfidence level is that the forecast interval of 1-a is $(C^T\widehat{\mathrm{\β}}\±t_{\frac{a}{2}}\widehat{\mathrm{\σ}}\sqrt{1+\frac{1}{n}+\frac{{(x_0-\stackrel{\‾}{x})}^2}{\mathrm{\Σ}{(x_i-\stackrel{\‾}{x})}^2}}).$

(4) relevant issues in the regretional analysis

1. multicollinearity problem

In multiple linear regression model, not linear dependence between each explanatory variable that important hypothesis is multiple linear regression, but actual when setting up multiple linear regression model, inevitably can introduce two or more explanatory variables, more or less exist interrelated between these variablees.

The consequence of ⅰ multicollinearity

Under perfect collinearity, (X ^TX) ^-1Do not exist,

Ask to come out, model is failed.

Under approximate collinearity,

| (X ^TX) ^-1| ≈ 0, so (X ^TX) ^-1It is very big to get cornerwise value, because

So the variance of the estimator of regression coefficient also can increase, the result can make the significance test inefficacy of variable, the forecast function of model weaken.

The check of ⅱ multicollinearity

Regression model method of inspection: with each x _iRemaining variables is returned, calculate the corresponding coefficient of determination

, set up statistic:

Obeying degree of freedom is the F distribution of k-2 and n-k-1, if F _iGreater than critical value, x then _iThere is collinearity with remaining variables.

Remedying of ⅲ multicollinearity

Mainly contain three kinds of ways: the variance of the Return Law, principal component analysis (PCA), minimizing estimator (increase sample size, merge variable) progressively.

The Return Law progressively: be explained variable with y, progressively introduce explanatory variable, constitute regression model, carry out model and estimate that if the goodness of fit is changed significantly, it is influential then to introduce variable, if it is not remarkable that the goodness of fit changes, then introduce variable and have collinearity with the variable of introducing earlier.

2. different variance problem

In multiple linear regression model, error term ε _iVariance and i irrelevant, namely

Be constant.But owing to the setting of model (the actual non-linear linearity that is set at), ignore important explanatory variable, the reasons such as measuring error of data,

Be different, because

Must be forbidden to make that the OLS estimation variance is being not minimum, the t check also no longer includes meaning.It is more that different variance occurs in cross-section data.

The check common method of different variance mainly contains G-Q check (variance Monotone Type, large sample, all satisfied except do not satisfy all the other conditions with the variance condition), White check, Glejser check, ARCH process (time series, large sample).

The White check is by an auxiliary regression formula structure χ ²Statistic is carried out different variance test.Be example: Y with the binary linear regression model _t=β ₀+ β ₁X _T1+ β ₂X _T2+ u _t

The null hypothesis H of White check ₀: the u in the following formula _tThere is not different variance, alternative hvpothesis H ₁: u _tThere is different variance.

Step1 at first carries out OLS to regression model and returns, and asks residual error e _t

Step2 does following regression model

$e_t^2=a_0+a_1{X}_{t1}+a_2{X}_{t2}+a_3{X}_{t1}^2+a_4{X}_{t2}^2+a_5{X}_{t1}{X}_{t2}+v_t$

Step3 asks the coefficient of determination, statistic

Degree of freedom k is the item number of explanatory variable in the auxiliary regression formula.

The decision rule of Step4White check is: if

, accept H ₀, if ${\mathrm{TR}}^2>{\mathrm{\χ}}_{a\left(k\right)}^2,$ Refusal H ₀

3. auto-correlation problem

In multiple linear regression model, E (ε ε ^T/ X)=σ ²I, but because the hysteresis quality of economic variable, the perhaps reasons such as specification error of model make E (ε _iε _j) ≠ 0, the correlativity of sequence namely takes place in i ≠ j, comprises that space correlation (cross-section data), time series are relevant.Covariance matrix D (ε)=σ of ε then ²Ω Ω ≠ I.Auto-correlation is the same with different variance can to make that the OLS estimation no longer is BLUE.

The autocorrelative check of ⅰ

Autocorrelation test has D-W method, Lagrange's multiplier check (LM check).

The Durbin-Watson statistic is weighed the first-order serial correlation of residual error, and computing method are:

$\mathrm{DW}=\stackrel{T}{\mathrm{\Σ}}{({\hat{u}}_t-{\hat{u}}_{t-1})}^2/\stackrel{T}{\mathrm{\Σ}}{{\hat{u}}_t}^2$

The scope of DW value is (0～4), and then there is not correlativity in sequence near 2, tables look-up according to the number of observed reading and judges whether sequence exists correlativity.

The DW statistic only is applicable to the single order autocorrelation test, and for high-order autocorrelation test and inapplicable.Utilize the LM statistic can set up an autocorrelation test method that applicability is stronger, both can check the single order auto-correlation, also can check the high-order auto-correlation.The LM check is finished by an auxiliary regression formula, and concrete steps are:

Step1 considers that error term is n rank autoregression form

u _t=ρ ₁u _T-1+ ...+ρ _nu _T-n+vtν _tBe random entry, meet various assumed conditionses.

The Step2 null hypothesis is H ₀: ρ ₁=ρ ₂=...=ρ _n=0

Step3 sets up residual error auxiliary regression formula:

$e_t={\hat{p}}_1{e}_{t-1}+...+{\hat{p}}_n{e}_{t-n}+{\mathrm{\β}}_0+{\mathrm{\β}}_1{X}_{1t}+{\mathrm{\β}}_2{X}_{2t}+...+{\mathrm{\β}}_k{X}_{\mathrm{kt}}+v_t$

Step4 calculates coefficient of determination R ², structure LM statistic:

LM＝TR ²

The progressive obedience of LM statistic Distribute, decision rule is: if

Accept H ₀

If $\mathrm{LM}={\mathrm{TR}}^2>{\mathrm{\χ}}_{\left(n\right)}^2,$ Refusal H ₀

The different variance of ⅱ and autocorrelative correction---Generalized Least Square GLS.

For having different variance and autocorrelative problem, available generalized linear model is represented:

y＝Xβ+ε D(ε)＝σ ²Ω

Wherein Ω is positive definite matrix, so Ω ^-1Also be positive definite, thereby have invertible matrix G, make Ω ^-1=G ^TG, so original linear model is carried out conversion:

Gy＝GXβ+Gε

So D (G ε)=σ ²G Ω G ^T, again because G Ω G ^T=I obtains D (G ε)=σ ²I.

In generalized linear model (1-19), there has not been the problem of auto-correlation and different variance like this:

$\widehat{\mathrm{\β}}={\left(X^T{\mathrm{\Ω}}^{-1}X\right)}^{-1}{X}^T{\mathrm{\Ω}}^{-1}y,$ The GLS of β estimates

Be the BLUE of β.

And σ ²Getting OLS estimates

Be σ ²Nothing partially, consistent Estimation.

Have only the multiple linear regression model by test of hypothesis just can be applied to practice.

3. nonlinear regression model (NLRM)

Taking the form of of the correlationship that exists between the independent variable of nonlinear regression model (NLRM) and dependent variable is nonlinear, often sees in the system of reality.But the nonlinear regression model (NLRM) complexity, operation easier is big, and common nonlinear model refers to that mainly those can pass through suitable substitution of variable, are converted into linear relationship with nonlinear relationship and handle.

(1) hyperbolic model:

(2) power function curve model: y=ax ^b(x＞0, a＞0)

(3) exponential model: y=ae ^Bx(a＞0)

(4) fall exponential model:

(5) sigmoid curve model:

3. the correlation analysis of steel industry electricity consumption

3.1 correlation analysis

Correlation analysis is the power of describing degree of correlation between the variable, and with the process that suitable statistical indicator shows, and its objective is to find out the factor that to steel industry electricity consumption has appreciable impact, as setting up model based.

3.1.1 related coefficient

Related coefficient also is Coefficient of production-moment correlation, is a common counter for quantitative description linear dependence degree quality.

According to as above supposing, establish (X ₁, X ₂) obey two-dimentional normal distribution

Wherein

Be X ₁Average, variance,

Be X ₂Average, variance, ρ is X ₁, X ₂Between related coefficient.The formula of related coefficient ρ is:

[0138] ρ＝Cov(X ₁,X ₂)/(Var(X ₁)Var(X ₂)) ^1/2 （5.1）

[0139]Utilize the square estimation technique, the square estimated value that obtains ρ is:

$r=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}(X_{1i}-{\stackrel{\‾}{X}}_1)(X_{2i}-{\stackrel{\‾}{X}}_2)}{{\left[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(X_{1i}-{\stackrel{\‾}{X}}_1)}^2\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(X_{2i}-{\stackrel{\‾}{X}}_2)}^2\right]}^{1/2}}---\left(5.2\right)$

3.1.2 stepwise regression analysis

In practical problems, people always wish to selecting some variablees as independent variable the influential all multivariates of dependent variable y, and the method for using multiple regression analysis sets up that " Optimal Regression Equation is in order to forecast or control dependent variable.It is so-called that " Optimal Regression Equation, mainly referring to wish to comprise in regression equation all influences significant independent variable to dependent variable y and does not comprise y is influenced inapparent independent variable.A kind of regression analysis that stepwise regression analysis puts forward according to this principle just.Its main thought be in whole independents variable of considering by it to the significance degree size of y contribution in other words, regression equation is introduced on descending ground one by one, and those may be introduced into regression equation all the time to the inapparent variable of y effect.In addition, the variable that oneself is introduced into regression equation also may lose importance after introducing new variables, and need reject away from regression equation.Introduce a variable or reject a variable from regression equation and all be called a step that progressively returns, each step all will be carried out the F check, to guarantee only to contain in the regression equation y is influenced significant variable before introducing new variables, and inapparent variable is disallowable.

The implementation process of stepwise regression analysis is all to calculate its sum of squares of partial regression (i.e. contribution) each step to the variable of introducing regression equation, select the variable of a sum of squares of partial regression minimum then, under F level given in advance, carry out significance test, if significantly this variable needn't from regression equation, reject, at this moment other several variablees do not need to reject yet in the equation.On the contrary, if not remarkable, then this variable will be rejected, and ascendingly successively other variable in the equation is carried out the F check by sum of squares of partial regression then.To influence inapparent variable to y and all reject, reservation all be significant.Then again the variable of not introducing in the regression equation is calculated its sum of squares of partial regression respectively, and select a wherein variable of sum of squares of partial regression maximum, under given F level, do significance test equally, if significantly then this variable is introduced regression equation, this process continues always, till when the variable in regression equation all can not reject and not have new variables and can introduce, at this moment progressively regression process finishes.

3.1.3 Granger (Granger) cause and effect check

The check of Granger cause and effect is by winner Clive Granger professor (CliveGranger) proposition of 2003 annual Nobel prize in economics and sets up.Than related coefficient and regretional analysis, the check of Granger cause and effect is separated causality with statistical angle from the cause-effect relationship of the priority sequential aspect explanatory variable of variation from correlativity.If again in conjunction with correct economic theory explanation, the economic dynamics of active agent relation in then reflecting reality more exactly.

Granger defines causality and has used the concept of information set, and has emphasized the sequential that event takes place.Make I _nBe all information in the universe till the n phase, Y _nBe all Y till the n phase _t(t=1,2 ... n), X _N+1Be the value of n+1 phase X, I _n-Y _nBe all information except Y.If the adding of Y has changed the probability distribution of X: F (X _N+1/ I _n) ≠ F (X _N+1/ (I _n-Y _n)), think that namely variable Y has the Granger causality to X.

In the reality, the distribution function of test variable is very difficult, and easier method is to handle from desired angle, check E (X _N+1/ I _n) ≠ E (X _N+1/ (I _n-Y _n)).If δ _N+1=E (X _N+1/ I _n)-E (X _N+1/ (I _n-Y _n)) be not 0 significantly, then Y is the Granger reason of X.Developed into afterwards with precision of prediction and checked the causality relation, if σ ²(X _N+1/ I _n)＜σ ²(X _N+1/ (I _n-Y _n)), then Y is the Granger reason of X.Information set I _nComprised that not only all correlated variabless have also comprised the unlimited lagged value of X and Y, but in reality, we can't obtain all data message I _n, can only be at obtainable information set J _nCondition under, the relation of variable is tested, this Granger cause-effect relationship is based on obtainable information set J _nDraw.

3.1.4 seasonal adjustment

To be Bureau of Census of the US Department of Commerce set up than the basis of method and grow up in moving average the X-12 method, it is characterised in that except adapting to the character of various economic targets, purpose according to various seasonal adjustments, outside the selection account form, under situation about not electing, also can be according to the statistics benchmark that enrolls in advance, feature by data is selected account form automatically, in computation process, can be according to the enchancement factor size in the data, adopt the moving average of different length, enchancement factor is more big, and moving average length is more big.This method has become a kind of quite meticulous, typical seasonal adjustment method through repeatedly adjustment and improvement, has been official and among the people, international body's employings such as (IMF) of states such as America and Europe, Japan, becomes the seasonal adjustment method of generally using at present.

3.1.5 trend adjustment

Tendency refers to time series X _tThe trend that has over time and change.Through the seasonal adjustment of X-12, we can access trend term and a fluctuation TC in the time series _t(TC _t=T _t+ C _t).In order to obtain the secular variation trend of constituent, need trend term is separated with the fluctuation item, fairly simple method such as regression analysis, exponential smoothing, method of difference are come the part of tendency in the processing time sequence, and that emphasis uses is HP(Hodrick-Prescott) filter method.

3.1.6 correlation analysis thinking

Correlation analysis has three purposes: the first, the influence factor that obtains in the qualitative analysis is quantized, and the index of influence factor can be quantitatively stated in establishment; The second, by the quantitative test of index being verified the correctness of qualitative analysis; The 3rd, three grades of influence factors are screened, obtain the important factor in order for analysis modeling.

Based on the above object, the method for using related coefficient, the check of Granger cause and effect and regretional analysis to combine is carried out correlation analysis.Three kinds of analytical approachs respectively have relative merits, adopt three kinds of quantitative analysis methods to carry out correlation analysis, can reach the purpose that complements each other, verifies mutually, improve the reliability of influence factor and correlation analysis.

The step of correlation analysis is:

(1) influence factor is quantized, determine influence factor index set.According to qualitative examination, analyze three grades of influence factors that draw the highly energy-consuming industry, understand domestic demand as the angle sorting from downstream industry, represent international demand from import and export amount and policy aspect.Describe real estate industry's demand with real estate investment, auto output is described automobile industry demand etc.

(2) come the linked character of initial analysis influence factor and power consumption with the trend comparison diagram of influence factor and power consumption.

(3) calculate the related coefficient of power consumption and product yield and influence factor, analyze their correlativity.

(4) adopt the method for Granger cause and effect check whether to have cause-effect relationship from statistical angle analysis power consumption and influence factor.

(5) adopt stepwise regression method that preceding two kinds of methods are selected influence factor analysis, reject unessential influence factor.

(6) on the basis as a result of comprehensive three kinds of analytical approachs, obtain the major influence factors of trade power consumption amount.

3.2 the correlation analysis of steel industry electricity consumption

3.2.1 influence factor index set

1, the analysis of index set and foundation

The influence factor of steel industry electricity consumption and relevant index gather for shown in the following table.

Table 5.1 steel industry electricity consumption influence factor and index summary sheet

2, the Collecting and dealing of data

The data owner that relates in the research will comprise electricity consumption data, industry data, product data.Wherein electricity consumption data comprise ferrous metal smelting and prolong and press the processing industry power consumption in year Dec in January, 2004-2009; Product data comprise crude steel, steel, pig iron production, and the downstream industry data owner will comprise product yield, the investment of each downstream industry, and the data such as domestic steel composite price index, prices of raw and semifnished materials index that are related to product margin.

Data Source mainly contains: national statistics board web, CEInet's staqtistical data base, China Iron ﹠ Steel Association's net, China Steel net, steel house net, search number net etc.The processing of data mainly comprises the processing of investment data, two aspects of processing of price data.Adopt prices for investment in fixed assets indexes (this season) to adjust to investment data, nominal investment data is converted to the actual investment data.The processing of price data mainly be will relevant price data be adjusted into be in January, 2004 benchmark decide the base data sequence.

3.2.2 correlation analysis result

1, to the The selection result of one-level influence factor

Analysis-by-synthesis by the whole bag of tricks, three indexs of product yield: crude steel output, steel output, pig iron output all with the power consumption height correlation, wherein the correlativity of steel output and power consumption is stronger, can consider when setting up forecast model with steel output as independent variable.

2, to the The selection result of two, three grades of influence factors

By related coefficient analysis, the check of Granger cause and effect, stepwise regression analysis three grades of indexs are screened layer by layer, can get correlation results and gather.Comprehensively screen through the whole bag of tricks, following index is the most remarkable to the influence of steel industry power consumption: investment in real estate development, transportation industry investment in fixed assets, auto output, domestic steel composite price index, prices of raw and semifnished materials index.

Table three is a grade influence factor correlation analysis result gather

4. the forecast model of steel industry electricity consumption

4.1 the regression model of steel industry power consumption and steel output

4.1.1 the regretional analysis of power consumption absolute number and steel output absolute number

If y _tBe the observed reading of power consumption, x _tObserved reading for steel output.At first adopt common least square (OLS) method solving model:

y _t＝c+βx _t+u _t （7.1）

Thereby obtain the essential information of model and corresponding coefficient check and model match situation.

1, the foundation of initial model and check

Steel industry power consumption and steel output are set up Linear Regression Model in One Unknown, the explanatory variable of model is steel output, explained variable is power consumption, the data that adopt are absolute number, the sample interval of data is year Dec in January, 2004-2009, totally 72 groups the moon degrees of data, the major parameter by Eviews software computation model and statistic check are shown in table 7.1.

The regression result of table 7.1 power consumption and output and relevant information

Table 7.1 has provided regression coefficient, the coefficient standard error of explanatory variable-steel output and constant term, and the t statistic of coefficient correspondence and followed probability thereof are used for the validity of test coefficient.If followed probability is effectively less than 0.05 explanation regression coefficient, is zero null hypothesis otherwise accept regression coefficient, show that this explanatory variable (or constant term) is uncorrelated with dependent variable.By preliminary regretional analysis, find that steel output has remarkable influence to the steel industry power consumption, the significance test of coefficient is passed through, and coefficient of determination value is 0.95, what the coefficient of determination reflected is the fitting effect of model, and this value illustrates that more near 1 the model match must be more good.

The actual value of power consumption and the trend of estimated value are closely similar, and the validity that model is estimated has been described; The mean absolute error of this model is 8.68 hundred million kilowatt hours, and average relative error is 3.77%.Through check, model exists auto-correlation and different variance.Therefore, also need initial model is adjusted.

2, to the adjustment of initial model

The method of lag period by adding correlative factor is come the autocorrelation of adjustment model, and corresponding regression result and relevant information are as shown in the table.

Table 7.2 adds regression result and the relevant information of lag period variable

Each index of the described presentation of results of last table has all been passed through significance test, and the goodness of fit value of model is 0.97, and fitting effect is very good.

Find that the autocorrelation of model has been eliminated in that above model is done in the further check analysis, but still have heteroscedasticity, and heteroscedasticity can exert an influence to the validity that model OLS estimates.Therefore on the basis of this model influence index, adopt weighted least-squares method to analyze, the regression result that obtains is as shown in the table.

The regression result of table 7.3 weighted least-squares method and relevant information

The described presentation of results model of last table has passed through whole conspicuousness and coefficient significance test, and the value of the goodness of fit simultaneously is 0.99, and fitting effect is very good.The mean absolute error of model is 8.11 hundred million kilowatt hours, and average relative error is 3.53%.

3, to the statistical test of model

(1) different variance test

In different variance test, the value of the corresponding followed probability of F statistic is 0.92, greater than 0.05, illustrates that there is not different variance in model.

Table 7.4ARCH (1) assay

(2) autocorrelation test

We adopt Ljun-Box Q statistic to come checking sequence relevant, and wherein the corresponding probable value of Q statistic of single order illustrates that all greater than 0.05 there is not serial correlation in the disturbance term of model arbitrarily.

(3) test of normality of residual error

The corresponding followed probability of described Jarque-Bera statistic is 0.72, greater than 0.05, the residual error item Normal Distribution of model is described, i.e. t statistic check is effective.

Comprehensive above check analysis illustrates that resulting model has satisfied the relevant precondition of least square method, and is effective during resulting regression model, is following form thereby the relation table of the absolute number of power consumption and steel output can be shown as:

y _t＝0.03x _t+0.42y _t-1+17.57 （7.2）

Wherein, y _tBe steel industry power consumption (hundred million kilowatt hours), x _tBe steel output (ten thousand tons), y _T-1Be preceding first phase steel industry power consumption.The implication of formula (7.2) is: the power consumption of steel industry is subjected to the influence of current steel output and preceding first phase power consumption, before known under the situation of first phase power consumption, and the steel output that every increase is 10,000 tons, the power consumption of steel industry will increase by 0.03 hundred million kilowatt hour.In conjunction with the analysis of front, the mean absolute error of model is 8.11 hundred million kilowatt hours, and average relative error is 3.53%.

4.1.2 the regretional analysis of power consumption speedup and steel output speedup

1, the foundation of initial model and check

Here use the of that month speedup on year-on-year basis of power consumption and steel output to set up initial regression model, the sample interval of data is year Dec in January, 2005-2009, totally 60 groups the moon degrees of data, set up regression model by Eviews software, regression result is shown in table 7.5.

The regression result of table 7.5 power consumption and output and relevant information

By preliminary regretional analysis, find that steel output speedup has certain influence to steel industry power consumption speedup, the coefficient significance test of steel output speedup is passed through.The goodness of fit value of model is 0.67, illustrates that fitting effect is general.And the average relative error of model is 6.7%, and error is bigger.Simultaneously, in conjunction with relevant check, also there is autocorrelation in model, therefore need adjust model.

2, to the adjustment of initial model

The method of lag period by adding correlative factor is carried out the adjustment of model, again model is carried out statistical test.Regression result and relevant information behind the adding lag period variable are as shown in the table.

Table 7.6 adds regression result and the relevant information of lag period variable

By model is carried out autocorrelative adjustment, the goodness of fit has had significant raising, illustrates that the fitting effect of model improves.Simultaneously, the whole conspicuousness of model and each coefficient significance test are all passed through.The fitting effect of the model that obtains is greatly improved, and estimated value and the trend map between the actual value of power consumption are closely similar, the average relative error 4.82% of model.

3, to the statistical test of model

(1) different variance test

In different variance test, the value of the corresponding followed probability of F statistic is 0.15, greater than 0.05, illustrates that there is not different variance in model.

Table 7.7ARCH (1) assay

(2) autocorrelation test

We adopt Ljun-Box Q statistic to come checking sequence relevant, and sample interval is year Dec in February, 2005 to 2009, and wherein the corresponding probable value of Q statistic of single order illustrates that all greater than 0.05 there is not serial correlation in the disturbance term of model arbitrarily.

(3) test of normality of residual error

The corresponding followed probability of described Jarque-Bera statistic is 0.38, greater than 0.05, the residual error item Normal Distribution of model is described, namely the t statistic check before is effective.

Comprehensive above check analysis illustrates that our resulting model has satisfied the relevant precondition of least square method, and is effective when namely regression model being described, is following form thereby the relation table of the speedup of power consumption and steel output can be shown as:

y _t＝0.55x _t+0.60y _t-1-0.13 （7.3）

Wherein, y _tBe the of that month speedup of steel industry power consumption, x _tBe the of that month speedup of steel output, y _T-1Be the of that month speedup of preceding first phase power consumption.

The implication of formula (7.3) is: the power consumption speedup of steel industry is subjected to the influence of current steel output speedup, preceding first phase power consumption speedup, before known under the situation of first phase power consumption, and the every increase by 1% of steel output, then the power consumption of steel industry increases by 0.55%.In conjunction with the analysis of front, the average relative error of model is 4.82%.

4.1.3 model comparative analysis

Power consumption and steel relation of yield model are contrasted as following table 7.8, two models have all passed through every check, illustrate that model is rationally, effectively, the error of fitting of two models is all smaller, form is also comparatively simple, comparatively speaking, the model error of fitting of prediction power consumption absolute number is littler, and fitting effect is better.

Table 7.8 power consumption and Relationship with Yield model gather

4.2 the regression model of steel industry power consumption and three grades of influence factors

In the correlation analysis between steel industry power consumption and the relevant downstream industry influence factor, we have obtained influencing comparatively significant six three grades of factors of steel industry power consumption: volume, transportation industry investment in fixed assets, automobile industry output, steel outlet amount, domestic steel composite price index, prices of raw and semifnished materials index are finished in investment in real estate development.For more effective power consumption to steel industry is predicted, below the relation between power consumption and influence index is carried out regretional analysis.

If y _tBe steel industry power consumption, x _1tFor volume is finished in investment in real estate development, x _1tBe transportation industry investment in fixed assets, x _3tBe automobile industry output, x _4tBe steel outlet amount, x _5tBe domestic steel composite price index, x _6tBe prices of raw and semifnished materials index.At first, adopt common least square (OLS) method solving model:

y _t＝c+β ₁x _1t+β ₂x _2t+β ₃x _3t+β ₄x _4t+β _5x5 _t+β ₆x _6t+u _t （7.4）

By to the finding the solution of above regression model, can obtain the actual value of power consumption and the match error of fitting of estimated value, thereby the validity of model is carried out preliminary judgement.Then, identical with the research of front, in polynary OLS estimates, need to return sequence equally and satisfy some preconditions, therefore also can do different variance, auto-correlation and test of normality to final model, could more effectively predict power consumption by the model of check.

4.2.1 the regression model of power consumption absolute number and three grades of influence factor absolute numbers

1, the foundation of initial model and check

Steel industry power consumption and three grades of influence factors are set up multiple linear regression model, the explanatory variable of model is various three grades of influence indexs, explained variable is power consumption, the data that adopt are absolute number, the sample interval of data is year Dec in January, 2004-2009, totally 72 groups the moon degrees of data, the major parameter by Eviews software computation model and statistic check are shown in table 7.9.

The regression result of table 7.9 power consumption and influence factor and relevant information

By preliminary regretional analysis, obtain the correlationship between power consumption and six three grades of influence factors, wherein except the transportation industry investment in fixed assets, the coefficient of other five influence factors has all passed through significance test, in order to guarantee the validity of model, after removing transportation industry investment in fixed assets index, obtained following regression result.

Table 7.10 power consumption and regression result and the relevant information of deleting the back influence factor

By further regretional analysis, five influence factors that obtain being left have very strong correlationship to power consumption, and corresponding coefficient has all passed through significance test, and the coefficient of determination value of model is 0.95, illustrate that fitting effect is very good, the significance test of model integral body is simultaneously passed through.The mean absolute error of model match is 9.52 hundred million kilowatt hours, and average relative error is 4.42%, and fitting effect is general.Further combined with the statistical test of model, model exists autocorrelation and heteroscedasticity.Therefore, also need to utilize relevant method, initial model is adjusted.

2, to the adjustment of initial regression model

Based on the analysis of front, by in model, adding the method for the preceding first phase of correlative factor, eliminate the autocorrelation of model, and it is as shown in the table to obtain corresponding regression result.

Table 7.11 adds regression result and the relevant information of lag period variable

By adjusting, all coefficients in the model have all passed through significance test, and the coefficient of determination value of model is 0.97, very good of the fitting effect that model is described.Model mean absolute error after the adjustment is 8.07 hundred million kilowatt hours, and average relative error is 3.82%, illustrates that our adjustment has reached the purpose that reduces model error.

3, to the statistical test of model

(1) different variance test

Table 7.12ARCH (1) assay

In different variance test, the value of the corresponding followed probability of F statistic is 0.32, greater than 0.05, illustrates that there is not different variance in model.

(2) autocorrelation test

We adopt Ljun-Box Q statistic to come checking sequence relevant, and sample interval is year Dec in February, 2004 to 2009, and wherein the corresponding probable value of Q statistic of single order illustrates that all greater than 0.05 there is not serial correlation in the disturbance term of model arbitrarily.

(3) test of normality of residual error

The corresponding followed probability of Jarque-Bera statistic is 0.72, greater than 0.05, the residual error item Normal Distribution of model is described, namely the t statistic check before is effective.

Comprehensive above check analysis illustrates that resulting model has satisfied the relevant precondition of least square method, and namely the regression model that obtains is effective, is following form thereby the relation table of the absolute number of power consumption and influence factor can be shown as:

y _t=1.18x _1t+0.48x _3t+5.10x _4t+0.46x _5t-0.90x _6t+0.43y _t-1+96.18（7.5）

Wherein, y _tBe steel industry power consumption (hundred million kilowatt hours), x _1tFor volume (10,000,000,000 yuan), x are finished in investment in real estate development _3tBe auto output (ten thousand), x _4tBe steel outlet amount (1,000,000 tons), x _5tBe domestic steel composite price index, x _6tBe prices of raw and semifnished materials index, y _T-1Preceding first phase for the steel industry power consumption.

The implication of formula (7.5) is: the power consumption of steel industry is subjected to current investment in real estate development to finish the influence of volume, auto output, steel outlet amount, domestic steel composite price index, the preceding first phase power consumption of prices of raw and semifnished materials exponential sum.And the relation of being proportionate between volume, auto output, steel outlet amount, domestic steel composite price index, the preceding first phase power consumption is finished in steel industry power consumption and investment in real estate development, and is negative correlativing relation between the prices of raw and semifnished materials index of steel industry.The mean absolute error of model is 8.07 hundred million kilowatt hours, and average relative error is 3.82%.

4.2.2 the regretional analysis of power consumption speedup and three grades of influence factor speedups

1, the foundation of initial model and check

Utilize the speedup on year-on-year basis of power consumption and three grades of influence factors to set up regression model below, the sample interval of data is year Dec in January, 2005-2009, totally 60 groups the moon degrees of data, set up regression model by Eviews software, regression result is as shown in the table.

The regression result of table 7.13 power consumption speedup and influence factor speedup and relevant information

The presentation of results of last table, the coefficient of the speedup of six influence factors has passed through significance test, and the coefficient estimated value that influence index namely is described all is effective.The value of the coefficient of determination is that the fitting effect of 0.86 explanation model is better.The average relative error of model is 4.12%.Further, in conjunction with relevant statistic check, there is autocorrelation in model.Therefore, the autocorrelation of model is adjusted.

2, do regretional analysis with generalized least square method

By using generalized least square method, the regression result that obtains model is as shown in the table.

The regression result of table 7.14 generalized least square method and relevant information

The coefficient check of the last table explanation prices of raw and semifnished materials is not passed through, though the fitting effect of model is fine, and check in conjunction with the ASSOCIATE STATISTICS amount, there is not autocorrelation in model under generalized least square method returns, but the speedup that still needs to remove prices of raw and semifnished materials index is carried out regretional analysis afterwards again, could guarantee the validity of the estimated result of model.

3, to the adjustment of initial model

Based on the analysis of generalized least square method, at first in initial regression model, remove this influence factor of prices of raw and semifnished materials speedup, regression result is as shown in the table.

Table 7.15 power consumption speedup and regression result and the relevant information of deleting back influence factor speedup

The described presentation of results of last table, except the investment in real estate development speedup, the coefficient of other indexs has all passed through significance test, therefore need further remove this index of investment in real estate development speedup.Secondly, because there is autocorrelation in model, but generalized least square method can not effectively reduce the error of fitting of model, therefore after we have eliminated prices of raw and semifnished materials index by generalized least square method, eliminated the investment in real estate development speedup again in conjunction with regretional analysis, then based on this in the model of table 7.15, add correlative factor early stage item method, eliminating the autocorrelation of model, and it is as shown in the table to obtain corresponding regression result.

Table 7.16 adds regression result and the relevant information of lag period variable

In the last table, each index has all been passed through significance test, and further combined with the ASSOCIATE STATISTICS check, model also exists heteroscedasticity.For the validity that guarantees that model is estimated, model is carried out carrying out new recurrence after different variance is handled, obtain the result of following table.

Table 7.17 is adjusted regression result and the relevant information of back model

The coefficient of described each influence factor of data declaration of last table has all passed through significance test, and the value of the coefficient of determination is 0.92, illustrates that the fitting effect of model is better.And the average relative error of this model is 3.54%, compares with initial regression model, and error reduces to some extent.

4, to the statistical test of model

(1) different variance test

Because the value of the corresponding followed probability of F statistic is 0.59, greater than 0.05, illustrates that there is not different variance in model.

Table 7.18 ARCH (1) assay

(2) autocorrelation test

Adopt Ljun-Box Q statistic to come checking sequence relevant, sample interval is year Dec in February, 2005 to 2009, and wherein the corresponding probable value of Q statistic of single order illustrates that all greater than 0.05 there is not serial correlation in the disturbance term of model arbitrarily.

(3) test of normality of residual error

The corresponding followed probability of Jarque-Bera statistic is 0.59, greater than 0.05, the residual error item Normal Distribution of model is described, namely the t statistic check before is effective.

Comprehensive above check analysis illustrates that resulting model has satisfied the relevant precondition of least square method, and is effective during the regression model that namely obtains, and is following form thereby the relation table of the speedup of power consumption and influence factor can be shown as:

y _t＝0.13x _2t+0.19x _3t+0.03x _4t+0.16x _5t+0.65y _t-1-0.11x _3t-1+0.12x _3t-2 （7.6）

Wherein, y _tBe steel industry power consumption speedup, x _2tBe transportation industry investment in fixed assets speedup, x _3tBe automobile industry output speedup, x _4tBe steel outlet amount speedup, x _5tBe domestic steel composite price index speedup, y _T-1Be the preceding first phase of steel industry power consumption speedup, x _3t-1Be the preceding first phase of automobile industry output speedup, x _3t-2Be preceding two phases of automobile industry output speedup, and each index is the of that month speedup based on raw data.

The implication of formula (7.6) is: the power consumption speedup of steel industry is subjected to the influence of current transportation industry investment in fixed assets speedup, auto output speedup, steel outlet amount speedup, domestic steel composite price index speedup and preceding first phase power consumption speedup, auto output speedup, preceding two phase auto output speedups.And the relation of being proportionate between steel industry power consumption speedup and transportation industry investment in fixed assets speedup, auto output speedup, steel outlet amount speedup, domestic steel composite price index speedup, preceding first phase power consumption speedup and the preceding two phase auto output speedups, and be negative correlativing relation between the preceding first phase auto output speedup.The average relative error of model is 3.54%, and fitting effect is better.

4.2.3 model comparative analysis

Two power consumptions in front and steel relation of yield model are contrasted as following table 7.19, two models have all passed through every check, illustrate that model is rationally, effectively, the error of fitting of two models is all smaller, form is also comparatively simple, comparatively speaking, the model fitting effect of prediction power consumption speedup is better.

Three grades of influence factor forecast models of table 7.19 gather

Claims (1)

1. the Forecasting Methodology of a steel industry electricity needs is characterized in that, this Forecasting Methodology comprises the following step:

(1) historical data of collection and arrangement steel industry power consumption and steel output;

(2) steel industry power consumption and steel output are set up Linear Regression Model in One Unknown, the explanatory variable of model is steel output, explained variable is power consumption, the The data absolute number, obtain regression coefficient, the coefficient standard error of explanatory variable-steel output and constant term, the statistic of coefficient correspondence and followed probability thereof are used for the validity of test coefficient; If followed probability is less than 0.05 then adopt regression coefficient, be zero hypothesis otherwise accept regression coefficient;

(3) method of the lag period by adding correlative factor is come the autocorrelation of adjustment model, again each index, integral body and coefficient is carried out significance test;

(4) model is carried out the test of normality of different variance test, autocorrelation test and residual error, make model satisfy the relevant precondition of least square method;

(5) pass that obtains the absolute number of power consumption and steel output by above-mentioned steps is:

Wherein,

Be the steel industry power consumption, unit: hundred million kilowatt hours;

Be steel output, unit: ten thousand tons,

Power consumption for preceding first phase steel industry;

(6) the of that month speedup on year-on-year basis of power consumption and steel output is set up regression model, the coefficient to steel output speedup carries out significance test and autocorrelation check again;

(7) method of the lag period by adding correlative factor is carried out the adjustment of model, model is carried out the test of normality of different variance test, autocorrelation test and residual error again, makes model satisfy the relevant precondition of least square method;

(8) obtain by above-mentioned steps and with the pass of the speedup of power consumption and steel output be:

Wherein,

Be the of that month speedup of steel industry power consumption, Be the of that month speedup of steel output,

Be the of that month speedup of preceding first phase power consumption.