Progressively Interactive Evolutionary Multi-Objective Optimization ...
Progressively Interactive Evolutionary Multi-Objective Optimization ...
Progressively Interactive Evolutionary Multi-Objective Optimization ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Difference in HV, DH(T)<br />
350000<br />
300000<br />
250000<br />
200000<br />
150000<br />
100000<br />
50000<br />
0<br />
−50000<br />
−100000<br />
A2<br />
x (terminated)<br />
−150000<br />
0 20 40 60 80 100 120 140<br />
Generation Counter (upper level), T<br />
Figure 32: Difference in hypervolume<br />
from ideal DH(T ) with upper level generation<br />
counter T for problem DS1 using<br />
three algorithms. Only algorithm A1 (H-<br />
BLEMO)reachesthe Pareto-optimalfront<br />
by making DH(T ) = 0.<br />
A3<br />
A1<br />
x<br />
x<br />
Difference in HV, DH(T)<br />
60000<br />
50000<br />
40000<br />
30000<br />
20000<br />
10000<br />
0<br />
A2<br />
x<br />
x<br />
x<br />
(terminated)<br />
−10000<br />
0 20 40 60 80 100 120 140 160<br />
Generation Counter (upper level), T<br />
Figure 33: Difference in hypervolume<br />
from ideal DH(T ) with upper level generation<br />
counter T for problem DS2 using<br />
three algorithm. Only algorithm A1 (H-<br />
BLEMO)reachesthe Pareto-optimalfront<br />
by making DH(T ) = 0.<br />
Table 3: Comparison of function evaluations for τ = −1 and τ = +1 cases with the<br />
H-BLEMOalgorithm.<br />
Prob. Best Median Worst<br />
No. Total LL Total UL Total LL Total UL Total LL TotalUL<br />
FE FE FE FE FE FE<br />
DS1(τ = +1) 2,819,770 87,582 3,423,544 91,852 3,829,812 107,659<br />
DS1(τ = −1) 3,139,381 92,624 3,597,090 98,934 4,087,557 113,430<br />
DS2(τ = +1) 4,484,580 105,439 4,695,352 116,605 5,467,633 138,107<br />
DS2(τ = −1) 4,796,131 112,563 4,958,593 122,413 5,731,016 144,428<br />
8 Scalability Study<br />
In this section, we consider DS1 and DS2 (with τ = 1) and show the scalability of our<br />
proposed procedure up to 40 variables. For this purpose, we consider four different<br />
variable sizes: n = 10, 20, 30 and 40. Based on parametric studies performed on these<br />
problemsinsection 6,we set Nu = 20n. Allother parametersareautomaticallyset ina<br />
self-adaptivemannerduring the course of asimulation, asbefore.<br />
Figure34showsthevariationoffunctionevaluationsforobtainingafixedtermination<br />
criterionon normalizedhypervolume measure(Hu < 0.0001)calculatedusing the<br />
upperlevelobjective valuesfor problemDS1. Sincethe verticalaxisis plotted inalogarithmic<br />
scale and the relationship is found to be sub-linear, the hybrid methodology<br />
performs better than an exponential algorithm. The break-upof computations needed<br />
inthelocalsearch,lowerlevelNSGA-IIandupperlevelNSGA-IIindicatethatmajority<br />
of the computations is spent in the lower level optimization task. This is an important<br />
insight to the working of the proposed H-BLEMOalgorithm and suggests that further<br />
efforts must be put in making the lower level optimization more computationally efficient.<br />
Figure 35 shows the similar outcome for problem DS2, but a comparison with<br />
that for problem DS1 indicates that DS2 is more difficult to be solved with an increase<br />
108<br />
A1<br />
A3