Multiscale island injection genetic algorithms for groundwater remediation

Genetic algorithms have been shown to be powerful tools for solving a wide variety of water resources optimization problems. Applying these approaches to complex, large-scale water resources applications can be difficult due to computational limitations, especially when a numerical model is needed to evaluate different solutions. This problem is particularly acute for solving field-scale groundwater remediation design problems, where fine spatial grids are often needed for accuracy. Finer grids usually improve the accuracy of the solutions, but they are also computationally expensive. In this paper we present multiscale island injection genetic algorithms (IIGAs), in which the optimization algorithms have different multiscale populations working on different islands (groups of processors) and periodically exchanging information. This new approach is tested using a field-scale pump-and-treat design problem at the Umatilla Army Depot in Oregon, USA. The performance of several variations of this approach is compared with the results of a simple genetic algorithm. The new approach found the same solution as much as 81% faster than the simple genetic algorithm and 9-53% faster than other previously formulated multiscale strategies. These findings indicate substantial promise for multiscale IIGA approaches to improve solution of complex water resources applications at the field scale.