is worth mentioning that normalized objective functions values are
used to calculate the performance metrics (Fig. 12).
The spacing metrics of obtained Pareto fronts of three algorithms
are shown in Fig. 13. From the box plot of spacing metric, it
can be observed that DMOPSO, MOGA, and MOPSO are capable of
providing competitive results in terms of this metric.
The result of maximum spread metric for the employed algorithms
is illustrated in Fig. 14. It can be noticed that DMOPSO
outperforms the other two algorithms. This measure indicates that
the generated Pareto front by DMOPSO algorithm covers a larger
area compared to the other algorithms.
As mentioned earlier, the true Pareto front of the problem is not
known. Consequently, the coverage metric is used to identify which
produced PF is closer to the true PF. The comparison results of the
coverage metric for three employed algorithms are presented in
Table 9. It indicates that most solutions laying over PF generated by
MOGA are dominated by the solutions found by DMOPSO. For
instance, 51% of solutions over PF resulted by MOGA are dominated
by the produced PF of DMOPSO algorithm while 2.1% of solutions of
DMOPSO Pareto front are dominated by solutions of MOGA. In
addition,14% of MOPSO solutions areworse than solutions found by
DMOPSO. In summary, the MOPSO has no problem in converging to
true Pareto front while it is difficult to cover a large area compared
to DMOPSO. The obtained PF resulted by DMOPSO indicates the
ability of this algorithm in dealing with the complex multiobjective
optimization problem. It can be concluded that the
generated PF by DMOPSO is more uniform and covers a larger area
than two other employed algorithms.