3. Numerical Implementation
For each test MOP, E-NSGA-II performs 10 runs with different seeds to observe the
consistency of the outcome. The mean value of the measures reveals the average
evolutionary performance of E-NSGA-II and represents the optimization results in
comparison with other algorisms. The variance of the best solutions in 10 runs indicates the
consistency of an algorithm. E-NSGA-II is implemented by MATLAB.
3.1 Performance Measures
Different performance measures for evaluating efficiency have been suggested in literature
(Okabe et al., 2003). For comparison purpose, this study applies two metrics: 1) the
convergence metric (Υ): approximating the average distance to the Pareto-optimal front; and
2) the diversity metric (Δ): measuring the extent of spread achieved among the obtained
solutions (Deb, 2001).
For the convergence metric (Υ), a smaller metric value implies a better convergence toward
the Pareto-optimal front. This study uses 500 uniformly spaced solutions to approximate the
true Pareto-optimal. To measure the distance between the obtained non-dominated front (Q)
and the set of Pareto-optimal solutions (P*), this study computes the minimum Euclidean
distance of each solution from 500 chosen points on the Pareto-optimal front by Equation (8).
The average of these distances is used as the convergence metric as Equation (9).