12.01.2015 Views

Download - Academy Publisher

Download - Academy Publisher

Download - Academy Publisher

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Theorem 1 [ 2]<br />

The necessary and sufficient condition of<br />

sequence X be homogeneous exponential function<br />

sequence isσ ( k)<br />

= const > 0,<br />

k ∈ K .<br />

Theorem 1 shows that sequence X would obey the<br />

white homogeneous index law when σ(k)<br />

is a fixed<br />

constant. Generally if σ (k ) is close to a fixed constant,<br />

we consider X has homogeneous grey index law.<br />

In actual problem, for the sequence which has<br />

homogeneous grey index law, it needs us to find a white<br />

~ ~ ~ ~<br />

−a k<br />

exponential function X = { x(<br />

k)<br />

= ce , k ∈ K}<br />

as the model<br />

of X frequently. From theorem 1, we can find that this<br />

method is equivalent to choosing a suitable value as the<br />

class ratio ~ σ , then determining the value of c ~ , in order to<br />

~<br />

let X have the properties of X . So we should choose<br />

suitable value of ~ σ and c ~ , then establishing optimized<br />

model.<br />

For example [ 2]<br />

, when we use grey model make midlong<br />

term prediction, we can let<br />

~<br />

n<br />

(1)<br />

σ = ∑σ<br />

( k)<br />

~<br />

~<br />

a = lnσ ,<br />

1<br />

1<br />

n − k=<br />

2<br />

n<br />

∑<br />

x<br />

~<br />

k = 1<br />

c =<br />

n<br />

∑<br />

k = 1<br />

~<br />

−a k<br />

( k)<br />

e ,and c ~ is the parameter which<br />

e<br />

~<br />

−2<br />

a k<br />

~<br />

makes error square sum between X and X be minimum;<br />

When we use grey model make short term prediction,<br />

we can let<br />

~<br />

n<br />

σ = 1<br />

∑ ξ<br />

kσ<br />

( k),0<br />

< ξ ξ ξ<br />

n<br />

ξ ξ ... ξ<br />

...<br />

2<br />

<<br />

3<br />

< (2)<br />

+ +<br />

2<br />

3<br />

n k=<br />

2<br />

~ ~ ~ ~<br />

~<br />

a n<br />

a = lnσ , c = x(<br />

n)<br />

e , and c is the parameter which<br />

~<br />

(<br />

makes x n)<br />

= x(<br />

n)<br />

.It further reflects the importance of<br />

new information.<br />

~<br />

Above method which establishing model X through<br />

determining development coefficient a is called class<br />

ratio modeling method. Especially, the method based on<br />

(1) is called average class ratio modeling method, and the<br />

method based on (2) is called weighted average class<br />

ratio modeling method.<br />

[2]<br />

Theorem 2 Average class ratio and weighted<br />

average class ratio modeling method are both conform to<br />

the homogeneous white index law.<br />

Theorem 2 shows that the class ratio modeling method<br />

has made up for the deficiency of GM (1, 1) modeling<br />

method because GM (1, 1) modeling method is not<br />

conform to the white homogeneous white index law.<br />

Obviously, the class ratio modeling method uses a<br />

~ ~ ~<br />

−a<br />

k<br />

white exponential function x(<br />

k)<br />

= c e to fit the sequence<br />

which has non-homogeneous grey index law. The key to<br />

it is finding a suitable value as the class ratio ~ σ , then<br />

determining the value of ~ a and ~ c .Reference [2] chooses<br />

value of the ~ σ based on average and weighted average<br />

of original sequence ratio. But any average value of class<br />

ratio in original sequence is always between the<br />

minimum ratio and the maximum ratio, then the class<br />

ratio modeling method proposed by reference [2] is not<br />

suitable for the non-steady primitive sequence which has<br />

non-homogeneous grey index law.<br />

III. WEAKENING CLASS RATIO MODELING METHOD<br />

A. Summary of the weakening class ratio modeling<br />

method<br />

We can use weakening buffer operator to affect the<br />

primitive behavior data sequence which possesses<br />

characteristics: the front part grows (weakens)<br />

excessively quickly and the latter part grows (weakens)<br />

excessively slowly. Its purpose is embodying the<br />

importance of new information, mitigate growth<br />

(weakens) speed of the front part data, and make the<br />

growth (weakens) speed of buffer sequence become<br />

steady. Then make it be easy to construct white<br />

~<br />

−a k<br />

exponential function x(<br />

k ) = c e and fit buffer<br />

sequence with it, thus it can improve prediction precision<br />

of model GM (1, 1). For the original sequence which<br />

should be pretreated by weakening buffer operator,<br />

author in this article introduces a new method of how to<br />

choose suitable class ratio and how to find class ratio<br />

modeling method on this kind of sequence directly.<br />

Theorem3 Let X = { x(<br />

k)<br />

x(<br />

k)<br />

> 0 or<br />

x ( k ) < 0, k = 1,2,3.... n}<br />

be the primitive behavior data<br />

sequence which possesses characteristics: the front part<br />

grows (weakens) excessively quickly and the latter part<br />

grows (weakens) excessively slowly, and<br />

XD = { x (1) d , x (2) d ,..., x ( n ) d } be the buffer sequence after<br />

X affected by weakening buffer operator . Then the<br />

back class ratio of the buffer sequence<br />

x(<br />

k −1)<br />

d<br />

{ σ ( k ) d σ ( k)<br />

d = , k = 2,3,...., n}<br />

is more close to 1<br />

x(<br />

k)<br />

d<br />

than the class ratio of the original<br />

sequence x(<br />

k − 1)<br />

{ σ ( k ) σ ( k ) = , k = 2,3,...., n<br />

x(<br />

k )<br />

} .<br />

Proof: 1) When { x ( k )} is monotone increasing<br />

about k , then the class ratio of each<br />

point x(<br />

k −1)<br />

σ ( k)<br />

= ≤ 1.<br />

x(<br />

k)<br />

And the growth speed of buffer sequence XD becomes<br />

steady after X affected by weakening buffer operator.<br />

We can get<br />

, then<br />

x ( k ) − x ( k − 1) ≥ x ( k ) d − x ( k − 1)<br />

d<br />

x(<br />

k)<br />

x(<br />

k)<br />

d<br />

≥ , namely x ( k − 1) x ( k − 1) d .<br />

≤<br />

x(<br />

k −1)<br />

x(<br />

k −1)<br />

d<br />

x ( k ) x ( k ) d<br />

So we can obtain σ ( k)<br />

≤ σ ( k)<br />

d ≤ 1 because the<br />

buffer sequence and the original sequence maintain the<br />

same monotony.<br />

2) When { x ( k )} is monotone decreasing about k ,<br />

then the class ratio of each point x(<br />

k −1)<br />

.<br />

~<br />

~<br />

σ ( k)<br />

= ≥ 1<br />

x(<br />

k)<br />

74

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!