Parallelization Strategies for Evolutionary Multiobjective Optimization

There has been a modern confluence of environmental and water systems research towards a design paradigm that emphasizes multiple objectives for computationally intensive applications in areas including groundwater management, water distribution systems, and non-point source pollution. To date there have been very few parallel evolutionary multiobjective applications world-wide. This study seeks to explore new parallelization strategies for the Epsilon-Dominance Nondominated Sorted Genetic Algorithm-II that compare standard Master-Slave and multiple population parallelization approaches. Our analysis focuses on enhancing the efficiency of evolutionary multiobjective optimization by (1) using convergence based dynamic topologies and (2) enhancing search progress using archive-based migration strategies. The long-term goal of this research is to develop parallelization strategies that minimize processor population sizes and communication times while maintaining a proper load balance to achieve optimal speedups. Detailed results of the analysis will be available at the time of the conference.