Date of Award
Summer 1997
Document Type
Dissertation - Restricted
Degree Name
Doctor of Philosophy (PhD)
Department
Civil, Construction, and Environmental Engineering
First Advisor
Novotny, Vladimir
Second Advisor
Feng, Xin
Third Advisor
Karshenas, Saeed
Abstract
Recent years have brought new or renewed research interest in several computational methods inspired by naturally occurring processes. These computational methods include genetic algorithms (GAs) and related methods of evolutionary computation, and artificial neural networks (ANNs). The area of engineering applications of GAs is relatively new and growing fast. On the one hand, GAs allow one to approach optimization of difficult problems to which conventional optimization methods cannot be applied. On the other hand, GAs are not likely to prove useful in areas where the traditional optimization methods work successfully. Global control of wastewater conveyance systems is a problem to which the traditional techniques have been applied and worked reasonably well, but required significant simplifications of flow routing. A formulation of an evolutionary approach, such as a GA, can be expected to allow the use of accurate flow routing models, including large unmodified simulation codes, because evolutionary computation requires no analytical expression or numerical approximation of derivatives. Investigation into the development of such a GA was the objective of the research described in this thesis. Additionally, the applicability of GAs to wastewater systems control can be enhanced if GAs are combined with ANNs. The ANNs' ability to provide nonlinear input-output mappings can be used to relax the time limitations on the GA search, because if trained to approximate a potentially slow GA, an ANN is likely to execute sufficiently fast and thus provide the final, on-line product. Combining a GA with an ANN for wastewater conveyance control was the second objective of this research.