HR: 17:15h
AN: H54B-06 [Abstracts]
TI: Comparing State-of-the-Art Evolutionary Multi-Objective Algorithms for Long-Term Groundwater Monitoring
Design
AU: * Reed, P M
EM: preed@engr.psu.edu
AF: The Pennsylvania State University, 212 Sackett Building, University Park, PA 16802
United States
AU: Kollat, J B
EM: juk124@psu.edu
AF: The Pennsylvania State University, 212 Sackett Building, University Park, PA 16802
United States
AB:
This study demonstrates the effectiveness of a modified version of Deb's Non-Dominated Sorted Genetic Algorithm II (NSGAII),
which the authors have named the Epsilon-Dominance Non-Dominated Sorted Genetic Algorithm II (Epsilon-NSGAII), at solving a
four objective long-term groundwater monitoring (LTM) design test case. The Epsilon-NSGAII incorporates prior theoretical
competent evolutionary algorithm (EA) design concepts and epsilon-dominance archiving to improve the original NSGAII's
efficiency, reliability, and ease-of-use. This algorithm eliminates much of the traditional trial-and-error parameterization
associated with evolutionary multi-objective optimization (EMO) through epsilon-dominance archiving, dynamic population
sizing, and automatic termination. The effectiveness and reliability of the new algorithm is compared to the original NSGAII
as well as two other benchmark multi-objective evolutionary algorithms (MOEAs), the Epsilon-Dominance Multi-Objective
Evolutionary Algorithm (Epsilon-MOEA) and the Strength Pareto Evolutionary Algorithm 2 (SPEA2). These MOEAs have been
selected because they have been demonstrated to be highly effective at solving numerous multi-objective problems.
The results presented in this study indicate superior performance of the Epsilon-NSGAII in terms of the hypervolume
indicator, unary Epsilon-indicator, and first-order empirical attainment function metrics. In addition, the runtime metric
results indicate that the diversity and convergence dynamics of the Epsilon-NSGAII are competitive to superior relative to
the SPEA2, with both algorithms greatly outperforming the NSGAII and Epsilon-MOEA in terms of these metrics. The improvements
in performance of the Epsilon-NSGAII over its parent algorithm the NSGAII demonstrate that the application of
Epsilon-dominance archiving, dynamic population sizing with archive injection, and automatic termination greatly improve
algorithm efficiency and reliability. In addition, the usability of the algorithm is improved through the elimination of the
population sizing parameter, the replacement of runtime specification by more intuitive termination criteria, and the
addition of Epsilon-dominance archiving which eliminates unnecessary costs associated with computing at unnecessary levels of
precision. This study contributes a comprehensive assessment methodology for MOEAs using runtime visualizations and
additional end-of-run performance metric results. Moreover, this study demonstrates that the Epsilon-NSGAII developed by the
authors has tremendous potential as an efficient, reliable, and easy-to-use MOEA for water resources applications.
DE: 1831 Groundwater quality
DE: 1832 Groundwater transport
DE: 1847 Modeling
DE: 1848 Monitoring networks
SC: Hydrology [H]
MN: Fall Meeting 2005