Optimization of soil vapor extraction using simulated annealing and genetic algorithms

Releases of volatile organic compounds (VOCs) to soil can pose a significant threat to human health and the environment. In situ soil vapor extraction (SVE) is used to remediate VOCs in the vadose zone. Costs for SVE systems can be several thousands to several millions of dollars, and if the SVE remediation technique is ineffective, contamination in deep soil may never be completely remediated.; One method to determine the cheapest, effective SVE system design is to use SVE simulation models. When using these models, the task to find the best design may be tedious (if not impossible) considering all possible well locations, screen depths, and vapor extraction rates. This research investigates the interfacing of heuristic optimization methods with SVE simulation codes to help determine a probable lowest cost SVE design, in minimal time, while still obtaining the required technical efficiency.; A two-dimensional analytical solution for a single SVE well operating under steady-state flow conditions was modified mathematically and with several rule-of-thumb parameters to incorporate a cost component that reflects well screen depth and vapor extraction rate. Although this ad hoc simulator should not be used in practice, it provided a quick way to evaluate different optimization methods applied to non-linear SVE equations.; Simulated annealing (SA), genetic algorithm, and simulated annealing/genetic algorithm (SAGA) hybrid optimizers were linked separately to the simulator. Comparison of several runs of these optimizers determined that a SA variant resulted in lower SVE cost and number of simulations required to find that cost. Additionally, it was found that the hybrid had better results but required much more programming.; Knowledge gained with these optimization methods was applied to a SVE coupled airflow/vapor transport simulation model (VENT3D) commonly used in industry. Because of time constraints, optimization of VENT3D using the SA variant, and not the hybrid, was investigated. When applied to a benchmark problem documented in literature, it was determined that the SA variant had comparable results to three other published optimizers. Furthermore, if the SAGA hybrid is used instead of the SA variant, it appears the hybrid could provide better results.