2013
60 48 05
: PGS.
- 2013
, ,
,
.TS.
Hà Nội, tháng 12 năm 2013
-
Hà Nội, tháng 12 năm 2013
LI CM N 4
MC LC 5
DANH M VIT TT 7
DANH MNG BIU 8
DANH M TH 9
M U 10
1. ỌỀ 10
2. MM V CA LUN V 10
3. I TU 10
4. PHU 10
5. C TIN C 10
6. B CC CA LU 11
Chng 1. TNG QUAN V 12
1.1. M U 12
1.2. C HIN D 13
1.3. PH 14
1.3.1. Ph 14
1.3.2. Ph nh lng 15
1.3.3. Mt s
i tit 15
1.3.4. Ph dng mng no 17
1.3.5.
17
1.4. T QU D 19
1.5. KT LUN CHNG 1 20
ChU MNG NO 21
2.1. GII THIU MNG NO 21
2.1.1. Mng n 21
2.1.2. Lch s n mng nron 21
2.1.3. ng nron vn thng 23
2.2. NRON SINH HO 24
2.2.1. Nron sinh hc 24
2.2.2. No 25
2.3. MNG NO 28
2.3.1. ng no 28
2.3.2.
29
2.3.3. Thut hc trong mng nron 30
2.4. 33
2.4.1. p 33
2.4.2. p 35
2.4.3. Thuc theo phn ngc sai s 36
2.4.4. Mt s yu t nh hc theoBPA 40
2.4.5. Mt s v c dng mng MLP 41
2.5.
NG 2 44
Chng 3. D TH I S RON
45
3.1. 45
3.2. U KIN NHI TH I 45
3.3. THCH D LIU 46
3.1.1. Th liu 46
3.1.2. ch d liu 47
3.4. D S D
48
3.4.1. 48
3.4.2. nh cng 49
3.4.3. Ch
52
3.4.4.
54
3.4.5. D
quy 55
3.5. D S D
58
3.5.1. D
-NN 58
3.5.2. D
61
3.6.
NG 3 64
KT LUN 65
1.
C 65
2. H
65
IU THAM KHO 66
1:
68
2: D
70
TT
1
ANN
Artificial Neural Network
Mnhn to
2
BPA
Back Propagation Algorithm
3
DWD
Deutscher Wetter Dienst
4
HRM
High Resolution Regional
Model
5
K-NN
K-Nearest Neighbors
K -
6
MAE
Mean Absolute Error
7
MAPE
MeanAbsolutePercentage
Error
Sai shn trm
tuyt
8
MLP
Multilayer Layer Perceptron
M nhiu
9
MSE
Mean Square Error
10
MM5
Fifth-Generation Penn
State/NCAR Mesoscale Model
Penn State /NCAR
5
11
RAMS
Regional Atmospheric
Modelling System
12
RMSE
Root Mean Square Error
13
WRS
The Weather Research And
Forcast
14
27
46
47
3.3
2010 49
3.4 2011 50
3.5
54
3.6
2010 57
3.7
-NN 60
3.8
63
-NN 17
25
H2.2 26
29
2.4
30
31
34
35
37
41
3.1 47
3.2
52
3.3
52
53
53
54
3.7
2010
55
3.8
2011
55
3.9 Kho 59
60
p K-NN 60
3.12
61
3.13
62
3.14
63
1.
tai, , .
2.
,
3.
,
,
.
-NN trong
.
.
,
,
2002 2011.
4.
5.
.
.
quy.
.
,
.
:
(
)
.
:
,
.
6.
- Chương 1: ,
.
- Chương 2:
- Chương 3:
-NN,
. , 3
.
.
.
1.1.
, , quy
,
.
,
,
,
,
.
:
-
-
-
-
-
-
-
, .
3 ,
,
.
2 ,
1.2.
-
-
-
-
-
,
,
1.3.
1.1
TT
1.
2.
olation
3.
4.
5.
-Impact Matrix Method
6.
7.
8.
1.3.1.
,
, ,
,
.
,
.
,
:
n
1.3.2.
quan,
.
-
-
,
1.3.3.
(High Resolution Regional Model) [7]:
-
Colorado
-04 do
, .
Arakawa-
1.3.4.
1.3.5.
1.3.5.1. Phương pha
́
p dư
̣
ba
́
o K-NN
K-Nearest Neighbors (K-NN)
.
-NN
,
-
,
(Query Point)
[24].
1.1,
(-,
1.1-NN
1.1
:
- 1 +
+.
- 2
+
kia.
- 5
-
.
1.3.5.2. Phương pha
́
p dư
̣
ba
́
o bă
̀
ng hô
̀
i quy
,
,
,
,
(
)
(
)
[9].
(1.1)
.
hay ph
.
1.4.
,
.
.
.
.
[18]:
- Sai số qun phương MSE(Mean Square Error)
(1.2)
- Sai số căn qun phương RMSE(Root Mean Square Error)
(1.3)
- Sai số tuyệt đối MAE (Mean Absolute Error)
(1. 4)
;
;
:
.
(1.2)
- -
(Coefficient of Determination )
.
(1.5) (1.6) [26]:
(1.5)
(1.6)
.
,
: ,
.
1.5.
.
,
2.1.
2.1.1.
Artificial NNetwork (ANN)
hay
2.1.2.
Organization of Behavior
g
Perceptron
-
.
Recognition
2.1.3.
2.2.
2.2.1.
, ,
. ,
.
4 :
- Cc nhnh v r: ,
.
+
, Na
+
hay Cl
-
.
,
, , .
.
(Weight).
- Thn th
̀
n kinh (Soma):
.
.
,
(
).
.
.
- Dy th
̀
n kinh (Axon): .
.
. .
- Khơ
́
p th
̀
n kinh (Synape): , n
. (2.1)
.
.
2.1
.
(Layer).
:
- Mng mt lp:
.
(Auto Associative).
- Mng hai lp:
.
- Mng nhiu lp:
.
(Hidden Layers).
- Mng truyn thng:
.
- Mng truyn ngưc:
.
- Mng t t chc :
(
).
, [14].
2.2.2.
.
,