NASA Technical Reports Server (NTRS) 20040171455: Using an Extended Kalman Filter Learning Algorithm for Feed-Forward Neural Networks to Describe Tracer Correlations

0.85 MB·English

by  NASA Technical Reports Server (NTRS)

Title: Using an Extended Kalman Filter Learning Algorithm for Feed-Foward
Neural Networks to Describe Tracer Correlations
Author(s): D. Law and Y. Mussa
David John Lary
Accepted by: Atmospheric Chemistry and Physics (ACP):
Reference: acpd-2004-0077
Neural networks are algorithms that attempt to mimic the brain by learning the
behavior of data. They can be thought of as a multivariate non-linear fit of inputs
to outputs.
In this study a new extended Kalman filter (EKF) learning algorithm for feed-
forward neural networks (FFN) is used. With the EKF approach, the training of
the FFN can be seen as state estimation for a non-linear stationary process. The
EKF method gives excellent convergence performances provided that there is
enough computer core memory and that the machine precision is high. Neural
networks are ideally suited to describe the spatial and temporal dependence of
tracer-tracer correlations. The neural network performs well even in regions
where the correlations are less compact and normally a family of correlation
curves would be required. For example, the CH4-N20 correlation can be well
described using a neural network trained with the latitude, pressure, time of year,
and CH4 volume mixing ratio (v.m.r.). The neural network was able to reproduce
the CH4-N20 correlation with a correlation coefficient between simulated and
training values of 0.9997.
n
e
0
a
*
0
0
cu
0
m..
U
0
0
(v
m
cn
J
B

Lo
Lo
(D
c>
r0 vr) N0 Lcoy
a,
ca.
PL cv)
a
av)
U
E
a
-cr
u)
S
8
a,
-Q
z
Z
LL
2
LL
Y
w
w-
0
c
K
E
-h
0
E
w
I I
0N INn
i
0
cu.
a>,
h4-adr
F
a
I-
O
Ui
E
.-
v)
a,
U i
0
U- c0
ra, a>,
i
% .%- c0
a0 2 0a>,
Y s
c c .-
3
cn30 aEE, ci00a>, 0cccnn
t 3cn 8
$ 4-
cEd 3Q 2
YC .K- Q
5a, 5a, sa, 5a,
.v-) .0- .v-) .v-)
xi
+ii 2 tiii
I I I I
d-
0
0
N
3
c3
w
0
LL
Y
w
w
.c-
rA
3
2
3
-a
a,
0
2
Q
w
.-
.c-
a
I=
4
cu
.-
0 0N
4
m
a,
c
v)
0
c
Y
0
a
n
0
u)
a,
.-
vz)
a,
r,
0
9 .+-
v
L
0
.c c-
!i
v)
a,
-
3 .-
a v)
> .a-
E L
.a-,
0 4-
U L
K 0
E .u- )
t
a, n
00 Q
60 .aac,. c0va,)
U s
I Q
a- 3 2=
ci cd r;
.-
z
!n v) 0N

T-
ci