Comparison of Optimization Methods for Ground-Water Bioremediation

Abstract

This paper compares computational performance of eight optimization algorithms used to identify the most cost-effective policy for in situ bioremediation of contaminated ground water. Six of these methods have not previously been applied to optimization of groundwater remediation. Numerical results were obtained for bioremediation of three problems based on two aquifers with time-invariant or time-varying pumping rates. Three major classes of algorithms are considered in the comparison: evolutionary algorithms [binary-coded genetic algorithm (BIGA), real-coded genetic algorithm, and derandomized evolution strategy (DES)], direct search methods (Nelder-Mead simplex, modified simplex, and parallel directive search), and derivative-based optimization methods (implicit filtering for constrained optimization and successive approximation linear quadratic regulator). Based on the three problems considered, the successive approximation linear quadratic regulator is the fastest algorithm. No one algorithm was consistently the most accurate on all three problems. The DES displayed an impressive combination of speed and accuracy. The DES has the advantage that it does not require derivative information. The BIGA was much slower and less accurate than all the other algorithms on the two problems BIGA solved.