Efficient Strategies for Sampling Uncertain Parameters in a Genetic Algorithm-Based Chance-Constrained Watershed Water Quality Management Problem

Chance-constrained genetic algorithms (CCGAs) can identify efficient solutions to complex water resources and environmental problems that satisfy a criterion with a specified likelihood under uncertainty. However, because they require evaluation of potential solutions numerous times using realizations of uncertain parameters, the computational time requirements can be very large. This paper investigates the effectiveness and efficiency of various strategies for sampling uncertain parameters in a CCGA applied to a watershed management design problem. Various sample sizes, as well as strategies for sampling uncertain parameters in a CCGA, such as using a new set of realizations for each generation, using the same set of realizations for all generations, using a new set of realizations for each potential solution evaluation, and strategies that vary the sample size based on the genetic algorithm (GA) progress, are investigated for a chance-constrained watershed management design problem. The resulting solution quality and computational requirements are compared, and recommendations are made for the most promising sampling strategies and sizes for GA-based chance-constrained watershed management.