Evolutionary Multiobjective Optimization in Watershed Water Quality Management

The watershed water quality management problem considered in this study involves the identification of pollution control choices that help meet water quality targets while sustaining necessary growth. The primary challenge is to identify nondominated management choices that represent the noninferior tradeoff between the two competing management objectives, namely allowable urban growth and water quality. Given the complex simulation models and the decision space associated with this problem, a genetic algorithm-based multiobjective optimization (MO) approach is needed to solve and analyze it. This paper describes the application of the Nondominated Sorting Algorithm II (NSGA-II) to this realistic problem. The effects of different population sizes and sensitivity to random seed are explored. As the water quality simulation run times can become prohibitive, appropriate stopping criteria to minimize the number of fitness evaluations are being investigated. To compare with the NSGA-II results, the MO watershed management problem was also analyzed via an iterative application of a hybrid GA/local-search method that solved a series of single objective ɛ-constraint formulations of the multiobjective problem. In this approach, the GA solutions were used as the starting points for the Nelder-Mead local search algorithm. The results indicate that NSGA-II offers a promising approach to solving this complex, real-world MO watershed management problem.