Application of beam-based optimization 217
0
20
40
60
80
100
cases
-2.5
-2
-1.5
-1
f
min
(mA/min)
I
=0.002 uA, N=1000
N-M
RS
RCDS-Imat
RCDS
0
20
40
60
80
100
cases
-2.5
-2
-1.5
-1
f
min
(mA/min)
I
=0.002 uA, N=500
N-M
RS
RCDS-Imat
RCDS
Figure 8.7 The final loss rate achieved by the four methods: N-M simplex, RSimplex,
RCDS with unit vectors for the initial direction set (RCDS-Imat), and RCDS with
calculated conjugate direction set (RCDS), in simulation for 100 cases each. Loss rate
noise sigma is 0.03 mA/min (not included in the plots). Left: with 1000 evaluations;
right: with 500 evaluations.
0
200
400
600
800
1000
evaluation
0.4
0.5
0.6
0.7
x
-2
-1
0
objective (mA/min)
RCDS
-2
-1
0
objective (mA/min)
RCDS-Imat
0
200
400
600
800
1000
evaluation
0.4
0.5
0.6
x
-2
-1
0
objective (mA/min)
RSimplex
0
200
400
600
800
1000
evaluation
0.4
0.5
0.6
x
-2
-1
0
objective (mA/min)
N-M Simplex
0
200
400
600
800
1000
evaluation
0.4
0.5
0.6
x
Figure 8.8 The convergence history in simulation for the case with a final minimum
ranking the 30th (out of 100 runs) for the four methods: RCDS with conjugate
direction set (top left), RCDS with unit vectors (top right), RSimplex (bottom left),
and N-M simplex (bottom right).