Major: Geological Engineering
Department: Geology and Geological Engineering
Mentor/Advisor: Dr. Liangping Li
Well Placement Optimization by Extremal Optimization
Author: Fleford Redoloza, Department of Geology and Geological Engineering
Mentor: Dr. Liangping Li, Department of Geology and Geological Engineering
The City of Aberdeen, South Dakota is planning to expand its current water supply. In preparation for future droughts and a growing population with increasing demands for water, the city is reassessing its water resources and is planning the construction of new wells to help meet growing water demands. To help the city reassess its water resources, the U.S. Geological Survey (USGS), in cooperation with the City of Aberdeen, released a newer and more accurate groundwater flow model for a study area just north of Aberdeen. To determine where to install future source wells, this model will be coupled with an optimization algorithm to approximate the best well field configuration that respects constraints such as costs, distance, drawdown, and river depletion. Current methods for well placement optimization are based on popular heuristics such as genetic algorithms and differential evolution. Majority of these algorithms rely on many evaluations of the groundwater model to evolve a wellfield solution. If the groundwater model is computationally intensive (like the Aberdeen groundwater model), then the optimization algorithm requires much more time to converge to a solution. To handle this computational bottleneck, this study proposes a modified version of the Extremal Optimization algorithm to reduce the number of groundwater model evaluations required. The algorithm operates on a single solution which it improves by modifying its worst performing components. The algorithm presented requires less function evaluations than popular methods because the fitness function used is structured to use more information from a single groundwater model evaluation. The proposed method was compared against differential evolution and genetic algorithms on a synthetic model. The proposed method was then applied to the Aberdeen groundwater model and constraint scenario.
Presentation Video