DIAGNOSTIC ASSESSMENT AND ADVANCEMENT OF MULTI-OBJECTIVE RESERVOIR CONTROL UNDER UNCERTAINTY

Abstract

This dissertation contributes to the assessment of new scientific developments for multi-objective decision support to improve multi-purpose river basin management. The main insights of this work highlight opportunities to improve modeling of complex multi-purpose water reservoir systems and opportunities to flexibly incorporate emerging demands and hydro-climatic uncertainty. Additionally, algorithm diagnostics contributed in this work enable the water resources field to better capitalize on the rapid growth in computational power. This opens new opportunities to increase the scope of the problems that can be solved and contribute to the robustness and sustainability of water systems management worldwide. This dissertation focuses on a multi-purpose reservoir system that captures the contextual and mathematical difficulties confronted in a broad range of global multi-purpose systems challenged by multiple competing demands and uncertainty. The first study demonstrates that advances in state of the art multiobjective evolutionary optimization enables to reliably and effectively find control policies that balance conflicting tradeoffs for multi-purpose reservoir control. Multiobjective evolutionary optimization techniques coupled with direct policy search can reliably and flexibly find suitable control policies that adapt to multi-sectorial water needs and to hydro-climatic uncertainty. The second study demonstrates the benefits of cooperative parallel MOEA architectures to reliably and effectively find many objective control policies when the system is subject to uncertainty and computational constraints. The more advanced cooperative, co-evolutionary parallel search expands the scope of problem difficulty that can be reliably addressed while facilitating the discovery of high quality approximations for optimal river basin tradeoffs. The insights from this chapter should enable water resources analysts to devote computational efforts towards representing reservoir systems more accurately by capturing uncertainty and multiple demands when properly using parallel coordinated search. The third study extended multi- purpose reservoir control to better capture flood protection. A risk-averse formulation contributed to the discovery of control policies that improve operations during hydrologic extremes. Overall this dissertation has carefully evaluated and advanced the Evolutionary Multiobjective Direct Policy Search (EMODPS) framework to support multi-objective and robust management of conflicting demands in complex reservoir systems.