Development of integrated simulation and optimization models for petroleum-contaminated groundwater remediation management under various uncertainties.

摘要

A number of integrated simulation-optimization models have been proposed in the past decades for groundwater remediation management. The major challenges in such efforts are how to alleviate computational costs in time-consuming simulation and optimization processes, and how to address various uncertainties encountered in modeling processes. In this dissertation research, four integrated simulation and optimization models were developed for optimal design of petroleum-contaminated groundwater remediation under various uncertainties. They are stochastic optimization model under parameter uncertainty (SOMUP), fuzzy optimization model under parameter uncertainty (FOMUP), stochastic optimization model under residual uncertainty (SOMUR), and groundwater optimization model under risk regulations (GOMUR).;The major contributions of this research are as follows. (1) A set of integrated models were developed for optimization of one surfactant-enhanced aquifer remediation (SEAR) and two pump-and-treat (PAT) systems under various uncertainties; these models could be extended to many other groundwater remediation systems, such as bioremediation, air sparging, soil vapor extraction, and dual-phase vacuum extraction, (2) As proxy simulators were used to replace the initial deterministic or nondeterministic numerical simulator, optimization times were reduced by many orders of magnitude compared to those without introducing proxy simulators. (3) A clusterwise linear regression (CLR) method was advanced; compared to traditional regression methods (e.g., stepwise cluster analysis), it has the advantages of avoiding the piecewise nature of prediction values, conducting finer analysis for differences not only between but also within clusters, and providing a reasonable results-interpretation mechanism. (4) A Monte-Carlo-based fuzzy simulation technique was proposed, which is useful for not only deriving possibilistic distributions of contaminant concentrations, but also for providing additional information such as the possibility of constraint satisfaction.;One laboratory-scale tank system and two field-scale sites were used as the study cases for demonstrating the practicability of developed models. Results from case studies indicated the models' capability to facilitate understanding the NAPL fate and transport in groundwater, answering questions such as what about the site situation in the future if a remediation action is or is not taken, handling complex uncertainties in stochastic or fuzzy environments, and identifying optimal groundwater remediation policies.