Keywords
Multi-Objective Optimization, Water Treatment, Climate Change, Visualization
Start Date
15-9-2020 7:20 PM
End Date
15-9-2020 7:40 PM
Abstract
The impact of climate change and increasing demand are projected to exacerbate water scarcity in many parts of the world. Impacts on the quality of drinking water sources, however, vary widely and are more difficult to quantify than quantity impacts. To deal with this uncertainty, water treatment simulation models enable planners to analyse operational and infrastructural alternatives across an ensemble of possible source water conditions. Manually developing these alternatives, however, does not guarantee that the best solutions are considered. To address this shortcoming, we developed a multiobjective optimization framework to generate treatment alternatives with robust performance across potential water quality futures. Using a multiobjective evolutionary algorithm coupled with a treatment model, we can examine the tradeoffs between competing treatment objectives, such as cost, disinfection byproduct formation, and the inactivation and removal of pathogens. To illustrate this technique, we studied a snowmelt dominated watershed in the western United States that has recently experienced wildfire and flooding events. These extreme events provide a proxy for the envelope of water quality uncertainty for that watershed. By creating an ensemble of water quality scenarios from the historical data, we developed an optimization problem to generate chemical dosing strategies for a conventional treatment plant in the area. We found that the plant would have the treatment capacity to handle increases in organic carbon up to 25% of historical peak concentrations. Because it is unlikely for organic carbon to reach these concentrations, the water utility would not need to consider infrastructural improvements to the plant in the near term. To visualize the optimization results, we use an open source interactive visualization library called Parasol, which can handle the large, high-dimensional data produced by multiobjective optimization.
Multiobjective Optimization to Support Long-term Water Treatment Planning
The impact of climate change and increasing demand are projected to exacerbate water scarcity in many parts of the world. Impacts on the quality of drinking water sources, however, vary widely and are more difficult to quantify than quantity impacts. To deal with this uncertainty, water treatment simulation models enable planners to analyse operational and infrastructural alternatives across an ensemble of possible source water conditions. Manually developing these alternatives, however, does not guarantee that the best solutions are considered. To address this shortcoming, we developed a multiobjective optimization framework to generate treatment alternatives with robust performance across potential water quality futures. Using a multiobjective evolutionary algorithm coupled with a treatment model, we can examine the tradeoffs between competing treatment objectives, such as cost, disinfection byproduct formation, and the inactivation and removal of pathogens. To illustrate this technique, we studied a snowmelt dominated watershed in the western United States that has recently experienced wildfire and flooding events. These extreme events provide a proxy for the envelope of water quality uncertainty for that watershed. By creating an ensemble of water quality scenarios from the historical data, we developed an optimization problem to generate chemical dosing strategies for a conventional treatment plant in the area. We found that the plant would have the treatment capacity to handle increases in organic carbon up to 25% of historical peak concentrations. Because it is unlikely for organic carbon to reach these concentrations, the water utility would not need to consider infrastructural improvements to the plant in the near term. To visualize the optimization results, we use an open source interactive visualization library called Parasol, which can handle the large, high-dimensional data produced by multiobjective optimization.
Stream and Session
false