Optimization and Machine Learning to Improve Water Resources Sustainability
Integrated water resources planning is an example of a large-scale, long-term infrastructure planning problem that is impacted by climate change and multiple potential design alternatives and goals. Traditionally, simulation models were used to evaluate a small number of alternatives in a benefit-cost analysis. Emerging studies of deep uncertainty, conditions where stakeholders do not know or cannot agree on the likelihood of uncertain events, have introduced new methods for water resources planning. These Decision Making Under Deep Uncertainty (DMDU) methods use many computer simulations with competing assumptions about uncertainty to 鈥渄iscover鈥 scenarios that cause vulnerabilities. This presentation will provide an overview of our research group鈥檚 DMDU work. Studies employ multi-objective evolutionary algorithm optimization, which develops tradeoff relationships between competing planning objectives; and machine learning, used to develop modeling relationships for inputs and outputs that do not have physically-based models. Examples include negotiations on water supply management in the Colorado River Basin, predicting drought in snow-dominated water supply systems, and predicting dam hazard.