- Research Article
- 10.1061/(asce)0733-9496(2007)133:6(566)
Closure to “Alternate Decision Making in Water Distribution Network with NSGA-II” by Md. Atiquzzaman, Shie-Yui Liong, and Xinying Yu
- Nov 01, 2007
- Journal of Water Resources Planning and Management
- Md Atiquzzaman + 2 more +2
The authors of the note used nondominating sorting genetic algorithms II NSGA-II for alternative decision making in a water distribution network. They coupled NSGA-II with water distribution network simulation software EPANET to obtain Paretooptimal front of the cost and the nodal pressure deficit. The discusser would like to mention the following points on the technical note. 1. The authors used NSGA-II for solving their multiobjective optimization problem. NSGA-II is an elitist nondominating sorting algorithm for solving multiobjective optimization problems Deb et al. 2000 . The efficiency of the algorithm is reported to be better than other multiobjective evolutionary algorithms in maintaining diversity among the solutions of the Pareto-optimal front. But this capability of maintaining diversity amongst the individual solutions on the Paretooptimal front is highly dependent on the scaling of the different objective functions Bhattacharjya 2004 . The discusser observed in Fig. 1 Fig. 2 of the technical note that the solutions are not evenly spread on the Pareto-optimal front. More concentration of the solutions is on the left side and very less concentration is on the right side of the plot. This uneven distribution of solutions on the Pareto-optimal front may be due to the improper scaling objective functions. It can be observed from Fig. 1 of the technical note that the range of first objective is in between 0 and 120 and that for the second objective is 200,000 and 450,000. In this situation, the NSGA-II will be biased toward the second objective without proper scaling of the objective functions. Therefore, it is necessary to scale down the second objective to the range of first objective or vice versa. The discusser believes that with proper scaling of the objective function, a better spread on the Pareto-optimal front can be maintained. Improper scaling may also lead to a nondominating front that is away from the true Pareto-optimal front. It is worth mentioning here that the recent version of the NSGA-II code, developed by Professor K. Deb and his students, has been modified incorporating normalized crowding distance criteria to maintain an even distribution of solutions on the nondominating front for problems having quite a different range of objective functions. However, it is always better to use proper scaling when the range of the objective functions is known. The authors show some solutions of the Pareto-optimal
Read more