Вы находитесь на странице: 1из 32

# $ %&&' ( #

# ) * %&&'

! "

# $ + ,,,

- .- - /0 12%3
C (t pj , o pj )
E p = C (t pj , o pj ) o pj = f j (net pj ) p w ji
w ji How to on Neural
C (t pj , o pj ) C (t pj , o pj ) net pj
w ji
=
net pj w ji Network Forecasting
C (t pj , o pj ) without Maths!
pj =
net pj
C (t pj , o pj ) C (t pj , o pj ) o pj
pj = =
net pj o pj net pj
o pt
= f j' ( net pj )
CD-Start-Up Kit for
net pj
C (t pj , o pj )
Neural Network
pj = f j' (net pj )
o pj
wki o pi
Forecasting
C (t pj , o pj ) net pk C (t pj , o pj )
net pk o pj
=
net pk
i
o pj
20+ software simulators
k k

C (t pj , o pj ) datasets
= wkj = pj wkj
net pk
k k
literature & faq
pj = f j' (net pj ) pj wkj
k

C (t pj , o pj ) ' 4 4 5 44 !
f ( net pj ) if unit j is in the output layer
o pj
j
pj = ,,, 4 6 # %&&'
f j' (net pj ) pk w pjk if unit j is in a hidden layer ,,,
k
# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* :5) 4;

# $ + ,,,

! " #$ %
Position: Senior Consultant Nationality: German Born: 1971

Languages Work Experience Method Competency

German 2004 - Research Associate, Lancaster SAP APO


English University, United Kingdom SAP Curriculum PLM100,
2000-2004 Research & Teaching Assistant SCM200, SCM 220
University of Hamburg, Prof. Inventory Management
Education
Premar & Vo ; Various Forecasting methods
2004 Post Doc Researcher at Int. projects in industry & trade Inventory Management
Centre for Forecasting, 1996-2000 CEO RSG Software GmbH. Supply Chain Planning/ APS
Lancaster University, UK Management & IT-Consultant Warehouse logistics
for Retail & Wholesale Projects Distribution- / Supply logistics
2003 Research Fellow at the in Germany & UK & Hungary. Business Information Systems
George Mason Uni, USA
Key-Projects
Problem Competency
2003 Visiting Scientist at the Forecasting Methods in SAP APO-DP,
Stellenbosch Business bdf HAM, Sales Forecast Management & Demand Planning in Consumer
School, USB, South Africa. bdf Netherlands Goods Industry
Automatic Model Selection in APO-DP, Demand Planning in Retail &
2000-2004 Research & Teaching bdf HAM, Sales Forecast Management Wholesale
Assistant University of Sales Forecasts, AOLTimeWarner, GER
Hamburg, PhD-thesis on Forecasting for Customer Relationship Member of the IBF Institute of
Forecasting in Inventory Management, Gruner & Jahr AG, GER Business Forecasters
Management using ANN Implementing a Forecast Strategy & Member of the International Institute
Inventory Management in Vending of Forecasters IIF
1992-1997 University of Hamburg Supply Chain, Mayfair Services, UK Member of IEEE; GOR, GI, ORSA
MBA equiv. (Dipl.-Kfm.) in Inventory Management in Distributions-
Business Administration Center Logistics, Vendepac, UK Regular Presentations at ANN
conferences of IEEE, INNS, APNNA

$ <'1 = '1 & && ( ! " 4 ( ! ,,, 4

%
# $ + ,,,

! & ' ()# * +

!
!

-
4 - ! 5 ) >7 7 ) ?@
, - ! )A )AB ;A A ?
>7 A C@
> - 4 , @
0 > - - , , -- 4 @
> 9- 7 7 @

- ! 6

4 !

# $ + ,,,

4 7 7 , .
;
A ,! B 7 7 , .
7 7 , . B0AB0)$ # 7
4 5 7 7 , .
7 7 , .
7 , .$
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

8
# $ + ,,,

) 7 7 , . >77@
D 4 4 4 , ,
- - . ?E , . ? - 4?
4 , 4 - F/ . 123

4 -
- - > @ , 9
4 4 /G 1=3

( ! -
- 4
n+1
4 - 4
4 >- 9 6 @ n +5
-
? n+ 2
-
n+6 - - -
n +3 ?
n+ h

n+ 4

# $ + ,,,

. "! $

77 # , 7 A 4

) A ; 9-
8 4 77! > % 2 @ % - 2 44 ( -
% - ? >- % @ - 4
< - 4

/ #
/# , 6 # 95H . 3
0)1 + $
0)

'
# $ + ,,,

. "! $

7 A 4 C D . 4 F

3 4$ 4

/ # $
-4 4 -

!2
- 5 -
4 I$

/
$
+ 3

# $ + ,,,

4 7 7 , .
;
A , 7 7 , .
7 7 , . B0AB0)$ # 7
4 5 7 7 , .
7 7 , .
7 , .$
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

*
# $ + ,,,

$
15667 & + 8
Citation Analysis by year
Forecast
title=(ne ural AND net* AND (forecast* OR pre dict* OR time-ser* OR
7 - time w /2 ser* OR timese r*) & title =(... )
Sales Forecast
Linear (Forecast)
5 9- and evaluated sales forecasting relate d point predictions
350 35
R2 = 0.9036
5 300 30
K > 7 ) @ 250 25
- 0 5 B- 0

[citations]
200 20

-- - 150 15
>;)0A)@ 100 10
9- 4 -
4 >7) ) 9- @ 50 5
5 4 H H >7) )@
0 0
6,

1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
0 - 4 [year]

7
Number of Publications by
- # Business Forecasting Domain
4 4 44
4 5

; ; 52 32
46; 4
83 General Business
6 9 Marketing
51
. Finance
Production
Product Sales
77 4 21
Electrical Load
;B ) 7C 10

# $ + ,,,

1J& 77 , 4 - 4 4 .
L B- M L )-- - 4

; ) 7 7 , . C
A - C

J
# $ + ,,,

( ' 9

Applied Forecasting Methods (all industries)


0% 10% 20% 30% 40% 50% 60% 70% 80%

Averages 25%

Autoregressive Methods 7%

Decomposition 8% $ 4
> O @ J P

[Forecastinng Method]
Exponential Smoothing 24%

Trendextrapolation 35%

Econometric Models 23%


4
Neural Networks 9%
> O @ %8P
Regression 69%

Analogies 23%

Delphi 22%
H 4 4
PERT 6%
> O @ %9P
Surveys 49%

[number replies]

* # %&&
%'& 8 4

77 -- 4 - ; 4A 6 5BA - C
/N ! 4 - M 3

# $ + ,,,

4 7 7 , .
;
A , 7 7 , .
7 7 , . B0AB0)$ # 7
4 5 7 7 , .
7 7 , .
7 , .$
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

=
# $ + ,,,

9 $ (-*-:;8

5 > @
) - 9 !- -
- - 4
- 4! 4 5 M 4 ,4I && %&&

5 - 9 - . , 4

# - >A @
A - 4 &8 >& %* M@ &1 >%*&& MA @
7 6$ & 5 &8 *& >A -@
N *&& . C
& J H &JH
-
- !K && - -

- ! I & &&& &&& &&& %& &&&

# $ + ,,,

<

; - 4 4 - 0 > 6A 1'8@
7 7 , .

-- 4

McCulloch / Pitts Rosenblatt Kohonen Rumelhart/Hinton/Williams


1943 1959 1972 1986
Hebb Minsky/Papert Werbos White
1949 1969 1974 1988

3
!

0 4 - 5)
7 A
#
4 KB )# )0Q! - CCC

2
# $ + ,,,

. /

; 77 -

#
- 7 4 4 - 4 6 4K
B - 7 4 > @ ; - 4 > @
$ A M
N A
? ?

4 R 4! ) SC

# $ + ,,,

4 7 7 , .
;
A , 7 7 , .
7 7 , . B0AB0)$ # 7
4 5 7 7 , .
7 7 , .
7 , .$
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

1
# $ + ,,,

9 $ (

, . ? , .

o1wi,1 neti = wijo j j ai = f (neti ) oi = ai

o1wi,1 o2 wi,2 oi
j


neti = wijo j ai = f (neti ) oi = ai

o2 oi,21woj wi,j
j

w j
oi
i,1 neti = wijo j j ai = f (neti ) oi = ai
w
oj oi,j2 wi,2 j
oi
o wi,j
j
n +1
n +5
n+2
n +6
n +3
n+h
9 + n+4

, oi = tanh w ji o j i
j
4,
neural_net = eval(net_name);
[num_rows, ins] = size(neural_net.iw{1});
[outs,num_cols] = size(neural_net.lw{neural_net.numLayers,
neural_net.numLayers-1});
if (strcmp(neural_net.adaptFcn,''))
net_type = 'RBF';
else net_type = 'MLP';
end

fid = fopen(path,'w');

# $ + ,,,

( ! = >
9 +
& && # 7 , &&&& #
- > @ A > @5B - > @7

- 3!/ ( !
- > - @
) > @
- T&E U
4 ! -
B - > 6 4 9 @
ui
o1 wi,1

wij o j j ai = f (net i )
wi,2
o2 neti = oi = ai oi
j

wi,j
oj
? aj f netj

0.5

oi = tanh w ji o j i , " 4 2 2 4
netj

0.5

j 1

&
# $ + ,,,

( ! " /
7 4 $ 0 B ; BV
, 4 4 -
% , 4 -
8 44
' - 4 /# 7)0Q V B ;C3
0 0 7, 9 - - ?
inputs weights information processing output

1%1
o1 w1,i
6%A0 1%1D6%A0
7@D 0%B@
7 0DC%60
w2,i 0%B@ ? )%101 = @%AAB>
@ o2 6%CCCB
? 6%CCCB oj
)%101
B
C%60 ? @%AAB
o3 w3,i
0

neti = wij o j j ai = f (neti ) oi = tanh w ji o j i


j j

# $ + ,,,

9 !
Architecture of a Multilayer Perceptron
Classic form of feed forward neural network!
Neurons un (units / nodes) ordered in Layers
unidirectional connections with trainable weights wn,n
Vector of input signals xi (input)
Vector of output signals oj (output)
w1,5 w5,8
X1 u1 u5 u8 w8,12
u12 o1 I , .
X2
u2 u5 u9
U13 o2
X3
u3 u6 u10
u14 o3

X4 u4 u7 u11

input-layer hidden-layers output-layer

ok = tanh wkj tanh wki tanh w ji o j j i k Min!


k i j
# $ + ,,,

4 7 7 , .
;
A , 7 7 , .
7 7 , . B0AB0)$ # 7
4 5 7 7 , .
7 7 , .
7 , .$
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

# $ + ,,,

/ !
$ )07 7V 0B W) A
M , 4 M 4, > . @
% , - A > 4 - 4 K @ >4 4 C@
8 , 4A - - -
' , 77 - 5 > 6 O @
* # ., 4 A - , -
0 0 7, 9 - - ?

w1,4 4 w4,9 Input-Vector Weight-Matrix

x1 x W
1 2 3 4 5 6 7 8 9 10
1 5 o1 1
9 2
x2 3
2 6 4
o2 5

x3 10 6
7
3 7 8
9
w3,8 w3,10 10
8
3?
o
Output-Vector

t
Teaching-Output

%
# $ + ,,,

/ ?3 9 E

M B7 , ,O
3= &>

4 4
- - %

:*- *
&

# $ + ,,,

3 ? ).7 : .

4
4
2
. 4 4 -
0

-2

-4
- M 9 4
, - 4
2
- I
- I4 -
0
, ,
,
4
-4 2
0

0
-2
-2 -4
6

5 4
0
2
2.5
0
2 0

-2.5

4 -4
-2 0
0
2
4

8
# $ + ,,,

. /

; 77 -

#
- 7 4 4 - 4 6 4K
B - 7 4 > @ ; - 4 > @
$ A M
N A
? ?

4 R 4! ) SC

# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
7 7 , . $ A 4
7 7 , . 6 $ A 4
7 7 , . - 7 0
4 - 77 4
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

'
# $ + ,,,

/ # !

)77 -- 9 / .6 6N 1% 3

-- > @ -- 9
- 4 > 4 @

y t + h = f ( xt ) + t + h

< I <
>@I 6
xt = vector of observations in t
et+h = independent error term in t+h

n +1
n +5
- - n + 2

4 F X n+ 6
n +3
C n+ h

- 4 5 n + 4

M O y t +1 = f ( yt , yt 1 , yt 2 ,..., yt n 1 )
$ = >

# $ + ,,,

/ / #

4 N 4 , )-- - ; (
!

-
( $

FG $
n+1

n+ 2
n +5
. (
n +3
#

n+ 4

*
# $ + ,,,

(
$
= >
(/<-H/
$ $

I (9
n+1
# + 4

8
n+2
n +5
3
n +3
?
n+4

y t +1 = f ( yt , yt 1 , yt 2 ,..., yt n 1 )
y t +1 = tanh wkj tanh wki tanh w ji yt j j i k
k i j
# $ = >

# $ + ,,,

/ # !

N - )77 5 9 - Y
/A J16V 4 2*3

? YYY
#

/ #

J
# $ + ,,,

/ # . ' / #

77 # , 7 A 4

9- !A 4 ) $ > @
$
$
$ 4$
$ 4, 44 $

# $ + ,,,

/ # !

N - )77 5 9 - Y
/A J16V 4 2*3

**
7B 4 - 4 - - 6
7B 4 - 4 B; $ B7 Z 4CCC
#( :*3 9-.3* !! - < 3 #( *38

=
# $ + ,,,

# '9 #

$
= >

n+1
n+ 5
n+ 2
n+ 6
n+ 3
n+ h

n+ 4

yt +1 , yt + 2 ,..., yt + n = f ( yt , yt 1 , yt 2 ,..., yt n1 )

# $ = >

# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
7 7 , . $ A 4
7 7 , . 6 $ A 4
7 7 , . - 7 0
4 - 77 4
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

2
# $ + ,,,

( $ 9

$
= >
3$

$
n +1

3
n+2
n +5 ?
n +3

n+4

yt +1 , yt + 2 ,..., yt + n = f ( yt , yt 1 , yt 2 ,..., yt n1 )

# $ = >

# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
7 7 , . $ A 4
7 7 , . 6 $ A 4
7 7 , . - 7 0
4 - 77 4
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

1
# $ + ,,,

9 9

(
#
9
9

n +1

n+2
n +5
n +3

n+4

y = f ( x1 , x2 , x3 ,..., xn )
y = x1 w1 j + x2t w2 j + x3t w3 j + ... + xn wnj j
9

# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
7 7 , . $ A 4
7 7 , . 6 $ A 4
7 7 , . - 7 0
4 - 77 4
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

%&
# $ + ,,,

$
# + 4

y t +1 = f ( yt , yt 1 , yt 2 ,..., yt n 1 )
yt +1 = yt wtj + yt 1 wt 1 j + yt 2 wt 2 j + ... + yt n 1 wt n 1 j j
$ = >

# $ + ,,,

= * >

3
y t +1 = f ( yt , yt 1 , yt 2 ,..., yt n 1 )
t n 1
yt +1 = tanh yi wij j
i =t $
$ = >

%
# $ + ,,,

* 9

y = f ( x1 , x2 , x3 ,..., xn )
y = x1 w1 j + x2t w2 j + x3t w3 j + ... + xn wnj j
* 9

# $ + ,,,

* 9

y t +1 = f ( yt , yt 1 , yt 2 ,..., yt n 1 )
t n 1
yt +1 = log yi wij j
i =t
9 =* > 9

%%
# $ + ,,,

9 +

9 - K 9 4
/&E 3E / 3
6 4 >& %? & 4 4 /&&& && & ?3
>& %8*E ?@
9 B - K
- 4 -
6 4 - 4 - -
>- - 4 @ B0 - -C
7 - K
?
7 B - K
?

- #( :*3 9 ;

# $ + ,,,

9
Forecasting Method

Objektive Subjektive Phrophecy


Forecasting Methods Forecasting Methods educated guessing
Time Series Causal
Methods Methods

Averages Simple Regression -


Moving Averages Multiple Regression )
Naive Mehods ; -
7 7 , .
Exponential Smoothing A 0$
Simple ES Z

Seasonal ES
; - 4$ 4
7 7 , . )0
Simple Regression 4
Autoregression ARIMA 4
7 7 , . 5 )7 4
) 5
Demand Planning Practice 0 ?
Objektive Methods + Subjektive correction

%8
# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
8 4 9 7 7 , .
9- )! $
9- #! $ 4
' $- 5 $ . - 7 7 , . 4
* : 5) , 4;

# $ + ,,,

/ # . * /

77 # , 7 A 4

9- ! 9 $ -- S M 0
8 4 77! > % 2 @ % - 2 44 ( -
; ! ? >- %
< - 4

/ #
/) 4 , 6 M 3
00@ + $
! ,B ,

%'
# $ + ,,,

/ # . ' 3$ 9

77 # , 7 A 4

9- ! ,
8 4 77! > % 2 @ % - 2 44 ( -
% - ? >- % @ - 4
< - 4

# 9 3$
$ *
! *

# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
$- 5 $ . ; A -
$- 5 $ . )
$- 5 $ . 7 , .$ 5
4 7 7 , .
* : 5) , 4;

%*
# $ + ,,,

. 9

; A -
7 M /&E 3 / E 3
4 77
7 7A $ 4
7 ;; 7 4
7 ;; 7 )Q 0
!

7 B $A $ 4
- 7 4 >) @
7 4 ; Z
$ 9- S , 4
9

M , > , Y@
$ 4> .- - 4 ?@
$ -
4 > -- @
)-- 7 7 , . 4

5 44

# $ + ,,,

/ / . ! !

; [ 4; \

B YQ C
; Y 7BC
; $ 4 6 Y 7B 6 4 - 4 6- - C

7 Y7 C
/&E 3 / E 3Y # BSC
)-- 4 YQ C
, 5- - Y 4
?

# 3

%J
# $ + ,,,

- ,

7 , . M 4
4
, 4 4 4
I - 4 Y

# 3

# $ + ,,,

/ / 9

; [ 4; \

7 - 4 Y ; A 7; C ) 6A) C
7 44 4 Y ; A 7; C > ,@
7 - 4 Y ; A 7; -- C

- 4 , . Y YYY
Y YYY

- Y$ CCC
- Y$ 4 C
?

# 3

%=
# $ + ,,,

/ / / #

; [ 4; \

4 , , K 4 Y 7B$ K) ;C
4 5 - Y ), - C
?

$ ) Y # .- - BS ;#; BS ?
A $ ) Y ; A 7; 4 C
-- Y Q ( , -- C
?

# 3

# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
$- 5 $ . ; A -
$- 5 $ . )
$- 5 $ . 7 , .$ 5
4 7 7 , .
* : 5) , 4;

%2
# $ + ,,,

# #

, A A ; ,

4 0 4
7 N . A
77
A
) - H77 H 77
4- HBB7
) 4 7 ?
7 A 4
7
7 A , 33 ; 0B
A 4 A ; 9-
8 -
0 4
94
0- . 4
7
,
?

/ 4< $ / + #

# $ + ,,,

# / # 8
) 4

N 4 )$ !7 A 4 7
V
X
, 4 7 % 7 $ 4 A ; $ 4
4
9- X

) A 4
A 4 A0B

A 4 4 #

7 A 77
77 A

7 ; 7
7 9
7
$ 4

%1
# $ + ,,,

# ':
7 , 7 N . A A

A A ;

) ) -

? ?

# $ + ,,,

* +
77 , ,,, ,,,
V , ) 77 , - ):
-!66 - 6- 6 6 ):
# Q ) #BBSCCC B YV !0 4 5 . R7 [

J
? 7 7 , . C
H# ( H #
H ( H
H (H

!
)
77 ( 7 7 , .
77 5 77 ( 5 - 7 7 , .

7 7 ! H 77 )77 5 B7 A > ?@
! # 5 C
7 , - , -
9- . ,? E@

8&
# $ + ,,,

4 7 7 , .
% )-- 7 7 , . #
8 4 9 7 7 , .
' $- 5 $ . - 7 7 , . 4
: 5) , 4;
)4 5; 4 7 7 , .
;

# $ + ,,,

$ $ . $ 8
$ . $

)77 )77
- > < C@ - > < C@
, - - , - -
4 4 4C 4 4 4C
)77 4 )77 4
4 4 4 4
4 , 9- C
4 0
- 4 0 - 9- 5 -
4 )77 , AB #
4
> C@
) 4
- 4
5 -
4 4
- - 4
?

8
# $ + ,,,

2 K ,

)77
- > < C@
, - -
4 4 4C
)77 4
4 4
9- C
0

" 4
N , !
#*(.3# ! !3 # $ + " 77 - O ,
,,, 4 - -- C
-- #)
,,, .
- O

# $ + ,,,

#$ %
0 )

; - 0 *'
) 'QW
4S 4

$ <'' >&@ *%' *182J=


$ <'' >&@ *%' *1812% 4
$ <'' >&@=2'& &J2 1
9 <'' >&@ *%' 2''22*

,,, .
" .

8%

Вам также может понравиться