CN103617458A - Short-term commodity demand prediction method - Google Patents

Abstract

The invention discloses a short-term commodity demand prediction method. The method comprises the steps that a historical data scatter plot is drawn according to the historical data of commodity flow, and selected abnormal values in the historical data are removed; a prediction model equation Y=gamma+beta*x+alpha*x<2> is built, the three-point method and the least square method are respectively used for calculating model parameters (alpha, beta, gamma), and the prediction model equations Y[1]=gamma[1]+beta[1]*x+alpha[1]*x<2> and Y[2]=gamma[2]+beta[2]*x+alpha[2]*x<2> are built respectively; the prediction model equations are used for calculating the predictive value at the tth period and calculating the difference of the tth period, the equation Y[2] is used for predicting the t+1th period if delta d[1] is larger than delta d[2], and the equation Y[1] is used for predicting the t+1th period if delta d[1] is smaller than delta d[2]; the selected prediction model equation is used for predicting the commodity flow at the t+1th period. The short-term commodity demand prediction method processes the influence caused by incomplete factor consideration and random disturbance terms on predicted results, only processes the data, and can precisely predict the recent data changing trend.

Description

A kind of short-term demand for commodity Forecasting Methodology

Technical field

The present invention relates to a kind of short-term demand for commodity Forecasting Methodology, be suitable for the historical data that current commodity have and also do not form the situation in a complete cycle, belong to information prediction technical field.

Background technology

Along with the lifting of worldwide production ability and the variation of consumer demand, the life cycle of commodity is shorter and shorter at present, so in a life cycle, the vital role of the variation tendency of commodity just seems and especially protrudes.Yet, the prediction to commodity flows Recent Changes trend, general businessman all by virtue of experience dopes a general numerical value to the commodity flows of lower first phase, or uses some forecast models.Yet first the forecast model of use be all adopt extract some on the larger factor of commodity flows impact as factor of influence, neglect the relatively little factor of commodity flows impact.Above these Forecasting Methodologies during due to deal with data distortion cause predicting the outcome accurate not.

Summary of the invention

Goal of the invention: for existing to the recent forward prediction out of true of commodity flows and model built-in problem, the invention provides a kind of short-term demand for commodity Forecasting Methodology, in the situation that do not consider to affect the factor of commodity flows and the weight of each factor impact, only data itself are processed, set up commodity flows database, not only commodity flows are carried out to qualitative forecasting and also commodity flows are carried out to accurate quantitative forecast, the flow information of commodity is provided to businessman in real time, thus for businessman provide clear, accurately, basis for estimation intuitively.

Technical scheme: a kind of short-term demand for commodity Forecasting Methodology, adopt curvilinear trend extrapolation method, specifically comprise the steps:

The first step: draw out historical data scatter diagram according to the historical data of commodity flows.

Second step: reject the exceptional value in selected historical data

1) calculate the mean value of sample data

variance with sample data $S^{}$ ${}^2=\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{-}{X})}^2,$ ? $S=\sqrt{\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{-}{X})}^2},$ X _ibe illustrated in i data value in time shaft ordered series of numbers

2) if

> 3S is x _ifor exceptional value, rejected, if

x _ifor normal value, retained.

The 3rd step: set up forecast model equation Y=γ+β x+ α x ²

The 4th step: use three point method and least square method to calculate respectively model parameter (α, beta, gamma)

Three point method:

1) get a little: when within the cycle, data amount check is less than 9, every group of three groups of data are got data, are respectively that first is first group of (x ₁, y ₁), middle one is second group of (x ₂, y ₂), last is the 3rd group of (x ₃, y ₃); When within the cycle, data amount check is more than or equal to 9, every group of three groups of data are got 3 data, are respectively that first three is first group of (x ₁₁, y ₁₁), (x ₁₂, y ₁₂), (x ₁₃, y ₁₃), middle three is second group of (x ₂₁, y ₂₁), (x ₂₂, y ₂₂), (x ₂₃, y ₂₃), last three is the 3rd group of (x ₃₁, y ₃₁), (x ₃₂, y ₃₂), (x ₃₃, y ₃₃); When within the cycle, data amount check is more than or equal to 15, every group of three groups of data are got 5 data, before five be first group of (x ₁₁, y ₁₁), (x ₁₂, y ₁₂), (x ₁₃, y ₁₃), (x ₁₄, y ₁₄), (x ₁₅, y ₁₅), middle five is second group of (x ₂₁, y ₂₁), (x ₂₂, y ₂₂), (x ₂₃, y ₂₃), (x ₂₄, y ₂₄), (x ₂₅, y ₂₅), last five is the 3rd group of (x ₃₁, y ₃₁), (x ₃₂, y ₃₂), (x ₃₃, y ₃₃), (x ₃₄, y ₃₄), (x ₃₅, y ₃₅).

2) assign weight: the weight allocation principle of every group of data is that the weight of a middle numerical value is ω ₀=3, toward both sides, assign weight as ω successively ₁=ω _-1=2, ω _-2=ω ₂=1

3) calculate the arithmetic mean of three groups of data computing formula is as follows:

$\left\{\begin{array}{c}\stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="1"\; label="Y"\; filenames="HDA0000432205400000052.png,HDA0000432205400000061.png"\; state="\{\{state\}\}">Y</figure-callout>}}_1}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}{Y}_{1i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}}=\frac{{\mathrm{\&omega;}}_{11}{Y}_{11}+{\mathrm{\&omega;}}_{12}{Y}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{1k}{Y}_{1k}}{{\mathrm{\&omega;}}_{11}+{\mathrm{\&omega;}}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{1\mathrm{<figure-callout\; id="2"\; label="k\; Y"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">k</figure-callout>}}}& \\ \stackrel{}{Y_{}}\stackrel{\&OverBar;}{{}_2}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}{Y}_{2i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}}=\frac{{\mathrm{\&omega;}}_{21}{Y}_{21}+{\mathrm{\&omega;}}_{22}{Y}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{+\mathrm{\&omega;}}_{2k}{Y}_{2k}}{{\mathrm{\&omega;}}_{21}+{\mathrm{\&omega;}}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{2\mathrm{<figure-callout\; id="3"\; label="k\; Y"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">k</figure-callout>}}}& \\ \stackrel{}{Y_{}}\stackrel{\&OverBar;}{{}_3}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{3i}{Y}_{3i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{3i}}=\frac{{\mathrm{\&omega;}}_{31}{Y}_{31}+{\mathrm{\&omega;}}_{32}{Y}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{3k}{Y}_{3k}}{{\mathrm{\&omega;}}_{31}+{\mathrm{\&omega;}}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{3k}}\end{array}\right.$

$\left\{\begin{array}{c}\stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="1"\; label="x"\; filenames="HDA0000432205400000052.png,HDA0000432205400000061.png"\; state="\{\{state\}\}">x</figure-callout>}}_1}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}{x}_{1i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}}=\frac{{\mathrm{\&omega;}}_{11}{x}_{11}+{\mathrm{\&omega;}}_{12}{x}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{1k}{x}_{1k}}{{\mathrm{\&omega;}}_{11}+{\mathrm{\&omega;}}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{1k}}\\ \stackrel{\&OverBar;}{x_2}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}{x}_{2i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}}=\frac{{\mathrm{\&omega;}}_{21}{x}_{21}+{\mathrm{\&omega;}}_{22}{x}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{+\mathrm{\&omega;}}_{2k}{x}_{2k}}{{\mathrm{\&omega;}}_{21}+{\mathrm{\&omega;}}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{2\mathrm{<figure-callout\; id="3"\; label="k\; x"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">k</figure-callout>}}}& \\ \stackrel{}{x_{}}\stackrel{\&OverBar;}{{}_3}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{3i}{x}_{3i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{3i}}=\frac{{\mathrm{\&omega;}}_{31}{x}_{31}+{\mathrm{\&omega;}}_{32}{x}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{3k}{x}_{3k}}{{\mathrm{\&omega;}}_{31}+{\mathrm{\&omega;}}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{3k}}\end{array}\right.$

ω _siit is the weight of i data in s group data;

4) calculate (α ₁, β ₁, γ ₁) formula is as follows

$\left\{\begin{array}{c}\stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="1"\; label="Y"\; filenames="HDA0000432205400000052.png,HDA0000432205400000061.png"\; state="\{\{state\}\}">Y</figure-callout>}}_1}={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_1{\stackrel{-}{x}}_1+{\mathrm{\&alpha;}}_1\stackrel{-2}{x_1}\\ \stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="2"\; label="Y"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">Y</figure-callout>}}_2}={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_1{\stackrel{-}{x}}_2+{\mathrm{\&alpha;}}_1\stackrel{-2}{x_2}\\ \stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="3"\; label="Y"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">Y</figure-callout>}}_3}={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_1{\stackrel{-}{x}}_3+{\mathrm{\&alpha;}}_1\stackrel{-2}{x_3}\end{array}\right.$

Least square method:

1) obtain historical True Data

2) calculate sum of squares of deviations, computing formula is as follows:

$Q=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{[y_t-{\hat{y}}_t]}^2=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{[y_t-({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)]}^2$

3) calculate (α ₂, β ₂, γ ₂) formula is as follows:

$\left\{\begin{array}{c}\frac{\&PartialD;Q}{{\&PartialD;\mathrm{\&gamma;}}_2}=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_t+\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)=0\\ \frac{\&PartialD;Q}{{\&PartialD;\mathrm{\&beta;}}_2}=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_t{x}_t+\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)x_t=0\\ \frac{\&PartialD;Q}{{\&PartialD;\mathrm{\&alpha;}}_2}\stackrel{}{=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_y{x}_{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">t</figure-callout>}}^2+\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)x_{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">t</figure-callout>}}^2=0}\end{array}\right.$

The 5th step: set up respectively forecast model equation Y ₁=γ ₁+ β ₁x+ α ₁x ², Y ₂=γ ₂+ β ₂x+ α ₂x ²

The 6th step: utilize forecast model equation to calculate the predicted value of t phase

and calculate the difference of t phase $\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="HDA0000432205400000052.png,HDA0000432205400000061.png"\; state="\{\{state\}\}">d</figure-callout>}}_1=|y_{1t}-{\hat{y}}_{1t}|,\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">d</figure-callout>}}_2=|y_{2t}-{\hat{y}}_{2t}|$

The 7th step: select forecast model equation.If Δ d ₁> Δ d ₂use equation Y ₂the t+1 phase is predicted, if Δ d ₁< Δ d ₂use equation Y ₁the t+1 phase is predicted

The 8th step: use the forecast model equation of selecting to predict t+1 phase commodity flows.

For the present invention, need following 2 points of explanation:

(1) this model is applicable to historical data and does not also form in the situation in complete cycle, if only need to extract the data in nearest one-period while being applied to know in the data analysis in cycle, but precision of prediction may there is no the precision of other forecasting techniques high.

(2), in three point method computation model parametric procedure, choosing of three points is that number according to sample data is selected.The sample data at most selection data of each point is just many, and sample data is few all includes all data with regard to as much as possible.

Beneficial effect: compared with prior art, ultimate principle of the present invention is by the matching to historical data data cycle of fluctuation, thus make development trend in this cycle and the accurately prediction of numerical value.The historical data that this Forecasting Methodology is suitable for having at present does not also form the situation in a complete cycle.In whole forecasting process, this Forecasting Methodology does not only consider to have how many kinds of factor more can not consider the weighing factor of each factor to predicted value to the impact of prediction effect actually.First, obtain historical data and using whole historical datas as sample data, draw out the scatter diagram of historical data.Then, rejecting abnormalities value, sets up curvilinear equation Y=γ+β x+ α x ², utilize historical data by three point method and least square method, to obtain equation coefficient (α, beta, gamma) respectively and set up respectively equation, dope respectively t phase commodity flows, obtain

poor subitem Δ d, select will with predictive equation the t+1 phase is predicted.The present invention is owing to not considering to affect factor and the weight of each factor on commodity flows impact of commodity flows, processed preferably because the inconsiderate complete and Disturbance of factor is on the impact predicting the outcome, only data itself are processed, can to recent data movement trend, be predicted more accurately.

Accompanying drawing explanation

Fig. 1 is the method flow diagram of the embodiment of the present invention;

Fig. 2 is the rejecting abnormalities value process flow diagram of the embodiment of the present invention;

Fig. 3 is that the estimates of parameters of the embodiment of the present invention is selected schematic diagram;

Fig. 4 be the embodiment of the present invention calculate model parameter process flow diagram;

Fig. 5 is the historical data curve map of the embodiment of the present invention;

Fig. 6 is the data and curves figure after the rejecting abnormalities value of the embodiment of the present invention;

Fig. 7 is the curve map after the prediction of the embodiment of the present invention.

Embodiment

Below in conjunction with specific embodiment, further illustrate the present invention, should understand these embodiment is only not used in and limits the scope of the invention for the present invention is described, after having read the present invention, those skilled in the art all fall within the application's claims limited range to the modification of the various equivalent form of values of the present invention.

As shown in Figure 1, short-term demand for commodity Forecasting Methodology, adopts curvilinear trend extrapolation method, specifically comprises the steps:

(1) obtain historical data, and draw scatter diagram, as shown in Figure 5;

i	1	2	3	4	5	6	7	8
									x i	1234	1256	600	1358	1425	1360	2000	1300

(2) reject the exceptional value in historical data, as shown in Figure 2:

1) sample average $\stackrel{\&OverBar;}{X}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i=1316.625,$ Sample standard deviation $S=\sqrt{\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{\&OverBar;}{X})}^2}=123.444536,$ 3S＝370.333608

2） $x_i-\stackrel{-}{X}$ Data are as follows

-82.625

-60.625

-716.625

41.375

108.375

43.375

683.375

-16.625

3) as seen from the above table $|x_3-\stackrel{\&OverBar;}{X}|>3S,|x_7-\stackrel{\&OverBar;}{X}|3S,$ So propose x ₃, x ₇

4) data after rejecting are as following table, and the curve map of drafting as shown in Figure 6.

X	1	2	3	4	5	6
							Y	1234	1256	1358	1425	1360	1300

(3) set up forecast model equation Y=γ+β x+ α x ²

(4) use three point method and least square method to calculate respectively model parameter (α, beta, gamma), as shown in Figure 4:

Three point method:

1) because sample data number is less than 9, so

$({\stackrel{-}{x}}_2,{\stackrel{-}{y}}_2)=\left(\mathrm{3,1358}\right),({\stackrel{-}{x}}_3,{\stackrel{-}{y}}_3)=\left(\mathrm{5,1360}\right)$

2) set up equation

$\left\{\begin{array}{c}1234={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_1+{\mathrm{\&alpha;}}_1\\ 1358={\mathrm{\&gamma;}}_1+3{\mathrm{\&beta;}}_1+9{\mathrm{\&alpha;}}_1\\ 1360={\mathrm{\&gamma;}}_1+5{\mathrm{\&beta;}}_1+25{\mathrm{\&alpha;}}_1\end{array}\right.$ Obtain $({\mathrm{\&alpha;}}_1,{\mathrm{\&beta;}}_1,{\mathrm{\&gamma;}}_1)=(-\frac{61}{4},\frac{492}{4},\frac{4505}{4})$

Least square method:

1） $\left\{\begin{array}{c}5{\mathrm{\&gamma;}}_2+15{\mathrm{\&beta;}}_2+55{\mathrm{\&alpha;}}_2=6633\\ 15{\mathrm{\&gamma;}}_2+55{\mathrm{\&beta;}}_2+225{\mathrm{\&alpha;}}_2=20320\\ 55{\mathrm{\&gamma;}}_2+225{\mathrm{\&beta;}}_2+979{\mathrm{\&alpha;}}_2=75280\end{array}\right.$ Obtain $({\mathrm{\&alpha;}}_2,{\mathrm{\&beta;}}_2,{\mathrm{\&gamma;}}_2)=(-\frac{2090}{140},\frac{18434}{140},\frac{153412}{140})$

(5) set up respectively predictive equation $Y_{}$ ${}_1={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_{1}x+{\mathrm{\&alpha;}}_1{x}^2=\frac{4505}{4}+\frac{492}{4}x-\frac{61}{4}{x}^2,$ ${\mathrm{<figure-callout\; id="2"\; label="Y"\; filenames="HDA0000432205400000051.png,HDA0000432205400000052.png"\; state="\{\{state\}\}">Y</figure-callout>}}_2={\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_{2}x+{\mathrm{\&alpha;}}_2{x}^2=\frac{153412}{140}+\frac{18434}{140}x-\frac{2090}{140}{x}^2$

The predicted value of (6) the 6th phases ${\hat{Y}}_{16}={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_{1}x+{\mathrm{\&alpha;}}_1{x}^2=\frac{4505}{4}+\frac{492}{4}\&times;6-\frac{61}{4}\&times;36=\frac{5261}{4}=1315.25$ ${\hat{Y}}_{26}={\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_{2}x+{\mathrm{\&alpha;}}_2{x}^2=\frac{153412}{140}+\frac{18434}{140}\&times;6-\frac{2090}{140}\&times;36=\frac{188776}{140}=1348.4$ ▽d ₁＝15.25，▽d ₂＝48.4

(7) because Δ d ₁< Δ d ₂so use equation $Y_{}$ ${}_1={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_{1}x+{\mathrm{\&alpha;}}_1{x}^2=\frac{4505}{4}+\frac{492}{4}x-\frac{61}{4}{x}^2$ The 7th phase was predicted

The predicted value of (8) the 7th phases is 1240, and the curve map after prediction as shown in Figure 7.

Claims (3)

1. a short-term demand for commodity Forecasting Methodology, is characterized in that, mainly comprises the steps:

(1) according to the historical data of commodity flows, draw out historical data scatter diagram;

(2) reject the exceptional value in selected historical data;

(3) set up forecast model equation Y=γ+β x+ α x ²;

(4) use three point method and least square method to calculate respectively model parameter (α, beta, gamma);

(5) set up respectively forecast model equation Y ₁=γ ₁+ β ₁x+ α ₁x ², Y ₂=γ ₂+ β ₂x+ α ₂x ²;

(6) utilize forecast model equation to calculate the predicted value of t phase and calculate the difference of t phase $\mathrm{\&Delta;}{d}_1=|y_{1t}-{\hat{y}}_{1t}|,\mathrm{\&Delta;}{d}_2=|y_{2t}-{\hat{y}}_{2t}|;$

(7) select forecast model equation; If Δ d ₁> Δ d ₂use equation Y ₂the t+1 phase is predicted, if Δ d ₁< Δ d ₂use equation Y ₁the t+1 phase is predicted;

(8) use the forecast model equation of selecting to predict t+1 phase commodity flows.

2. short-term demand for commodity Forecasting Methodology as claimed in claim 1, is characterized in that, the step of rejecting the exceptional value in selected historical data is:

1) calculate the mean value of sample data variance with sample data $S^2=\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{-}{X})}^2,$ ? $S=\sqrt{\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{-}{X})}^2}$

2) if

x _ifor exceptional value, rejected, if

x _ifor normal value, retained.

3. short-term demand for commodity Forecasting Methodology as claimed in claim 2, is characterized in that, the concrete steps of using three point method and least square method to calculate respectively model parameter (α, beta, gamma) are:

Three point method:

1) get a little: when within the cycle, data amount check is less than 9, every group of three groups of data are got data, are respectively that first is first group of (x ₁, y ₁), middle one is second group of (x ₂, y ₂), last is the 3rd group of (x ₃, y ₃); When within the cycle, data amount check is more than or equal to 9, every group of three groups of data are got 3 data, are respectively that first three is first group of (x ₁₁, y ₁₁), (x ₁₂, y ₁₂), (x ₁₃, y ₁₃), middle three is second group of (x ₂₁, y ₂₁), (x ₂₂, y ₂₂), (x ₂₃, y ₂₃), last three is the 3rd group of (x ₃₁, y ₃₁), (x ₃₂, y ₃₂), (x ₃₃, y ₃₃); When within the cycle, data amount check is more than or equal to 15, every group of three groups of data are got 5 data, before five be first group of (x ₁₁, y ₁₁), (x ₁₂, y ₁₂), (x ₁₃, y ₁₃), (x ₁₄, y ₁₄), (x ₁₅, y ₁₅), middle five is second group of (x ₂₁, y ₂₁), (x ₂₂, y ₂₂), (x ₂₃, y ₂₃), (x ₂₄, y ₂₄), (x ₂₅, y ₂₅), last five is the 3rd group of (x ₃₁, y ₃₁), (x ₃₂, y ₃₂), (x ₃₃, y ₃₃), (x ₃₄, y ₃₄), (x ₃₅, y ₃₅);

2) assign weight: the weight allocation principle of every group of data is that the weight of a middle numerical value is w ₀=3, toward both sides, assign weight as w successively ₁=w _-1=2, w _-2=w ₂=1;

3) calculate the arithmetic mean of three groups of data

computing formula is as follows:

$\left\{\begin{array}{c}\stackrel{\&OverBar;}{Y_1}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}{Y}_{1i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}}=\frac{{\mathrm{\&omega;}}_{11}{Y}_{11}+{\mathrm{\&omega;}}_{12}{Y}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{1k}{Y}_{1k}}{{\mathrm{\&omega;}}_{11}+{\mathrm{\&omega;}}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{1k}}\\ \stackrel{\&OverBar;}{Y_2}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}{Y}_{2i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}}=\frac{{\mathrm{\&omega;}}_{21}{Y}_{21}+{\mathrm{\&omega;}}_{22}{Y}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{+\mathrm{\&omega;}}_{2k}{Y}_{2k}}{{\mathrm{\&omega;}}_{21}+{\mathrm{\&omega;}}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{2k}}\\ \stackrel{\&OverBar;}{Y_3}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{3i}{Y}_{3i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{\mathrm{si}}}=\frac{{\mathrm{\&omega;}}_{31}{Y}_{31}+{\mathrm{\&omega;}}_{32}{Y}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{3k}{Y}_{3k}}{{\mathrm{\&omega;}}_{31}+{\mathrm{\&omega;}}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{3k}}\end{array}\right.$

$\left\{\begin{array}{c}\stackrel{\&OverBar;}{x_1}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}{x}_{1i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{1i}}=\frac{{\mathrm{\&omega;}}_{11}{x}_{11}+{\mathrm{\&omega;}}_{12}{x}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{1k}{x}_{1k}}{{\mathrm{\&omega;}}_{11}+{\mathrm{\&omega;}}_{12}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{1k}}\\ \stackrel{\&OverBar;}{x_2}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}{x}_{2i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{2i}}=\frac{{\mathrm{\&omega;}}_{21}{x}_{21}+{\mathrm{\&omega;}}_{22}{x}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{+\mathrm{\&omega;}}_{2k}{x}_{2k}}{{\mathrm{\&omega;}}_{21}+{\mathrm{\&omega;}}_{22}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{2k}}\\ \stackrel{\&OverBar;}{x_3}=\frac{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{3i}{x}_{3i}}{\mathrm{\&Sigma;}{\mathrm{\&omega;}}_{3i}}=\frac{{\mathrm{\&omega;}}_{31}{x}_{31}+{\mathrm{\&omega;}}_{32}{x}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&omega;}}_{3k}{x}_{3k}}{{\mathrm{\&omega;}}_{31}+{\mathrm{\&omega;}}_{32}+\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&omega;}}_{3k}}\end{array}\right.$

4) calculate (α ₁, β ₁, γ ₁) formula is as follows

$\left\{\begin{array}{c}\stackrel{\&OverBar;}{Y_1}={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_1{\stackrel{-}{x}}_1+{\mathrm{\&alpha;}}_1\stackrel{-2}{x_1}\\ \stackrel{\&OverBar;}{Y_2}={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_1{\stackrel{-}{x}}_2+{\mathrm{\&alpha;}}_1\stackrel{-2}{x_2}\\ \stackrel{\&OverBar;}{Y_3}={\mathrm{\&gamma;}}_1+{\mathrm{\&beta;}}_1{\stackrel{-}{x}}_3+{\mathrm{\&alpha;}}_1\stackrel{-2}{x_3}\end{array}\right.;$

Least square method:

1) obtain historical True Data and predicted data;

2) calculate sum of squares of deviations, computing formula is as follows:

$Q=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{[y_t-{\hat{y}}_t]}^2=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{[y_t-({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)]}^2;$

3) calculate (α ₂, β ₂, γ ₂) formula is as follows:

$\left\{\begin{array}{c}\frac{\&PartialD;Q}{{\&PartialD;\mathrm{\&gamma;}}_2}=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_t+\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)=0\\ \frac{\&PartialD;Q}{{\&PartialD;\mathrm{\&beta;}}_2}=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_t{x}_t+\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)x_t=0\\ \frac{\&PartialD;Q}{{\&PartialD;\mathrm{\&alpha;}}_2}\stackrel{}{=\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_y{x}_t^2+\underset{t=1}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&gamma;}}_2+{\mathrm{\&beta;}}_2{x}_t+{\mathrm{\&alpha;}}_2{x}_t^2)x_t^2=0}\end{array}\right..$

Images