A Self-adaptive Hybrid Genetic Algorihm for Optimal Groundwater Remediation Design By

Groundwater contamination is the result of multiple human activities, such as agriculture, industrial practices and military operations. The traditional remediation approach is to combine pump-and-treat for plume containment and contaminant capture, with other remediation technology for source control. Pump-and-treat systems are expensive, typically requiring high installation and operation costs. The traditional solution approach is then to proceed by trial-and-error, evaluating different design alternatives and selecting the best one from those evaluated. A large body of research has demonstrated that coupling optimization models with simulation models can aid in identifying effective remediation designs. The simple genetic algorithm (SGA) is a heuristic technique capable of solving these types of problems. Unfortunately, the solution of these complex problems is generally computationally intensive. This research focuses on the development and use of a hybrid genetic algorithm (HGA), a method that combines the use of SGA with local search to solve a groundwater remediation problem. The inclusion of local search helps to speed up the solution process and to make the solution technique more robust. The result of this research is a highly iii iv reliable numerical tool, the enhanced self-adaptive hybrid genetic algorithm (e-SAHGA) to more efficiently and effectively solve problems using simple genetic algorithms (SGAs). With this tool, the designer can evaluate different solution alternatives in a more timely fashion. A step-by-step methodology has also been developed for evaluating the optimal parameters for using the algorithm. This methodology, together with the adaptive nature of the algorithm, reduces the need for trial-and-error experiments to determine the optimal set of parameters for the algorithm. The application of the e-SAHGA algorithm to a hypothetical groundwater remediation design problem showed 90% reliability in identifying the solution faster than the SGA, with average savings of 64% across 100 runs with different random initial populations. Finally, e-SAHGA was tested on a field-scale remediation design problem, re-evaluation of the remediation system for Umatilla Army Depot, where it gave computational savings between 30% and 60% and, for one solution method, found a solution that was 4% better than the one found by the SGA. To my Mom and in loving memory of my Dad