Environmental Engineering Ph.D. candidate
What to Expect
In the age of abundant data, this resource will play an increasingly important role in the management and study of water systems. Like self-driving cars, we can imagine smart water systems that autonomously change their behavior based on real-time measurements to reduce flooding and improve water quality. However, reaching this vision poses a number of questions at the intersection of environmental and water resources engineering and data science, including (1) how to extract actionable information about system conditions from noisy and variable sensor data, (2) how to scale control to coordinate many distributed assets across a system, and (3) what trade-offs exist between objectives (e.g., minimize flooding, reduce overflows, maximize treatment) when making control decisions and how can a control algorithm be formulated to elucidate and balance these priorities. To address these questions, this talk will first present coupled signal processing and machine learning techniques in an automated toolchain to provide estimates of the current and future sewer dynamics from historical measurements. More broadly, it is identified that more data do not always result in a better model due to temporal shifts in system behavior and that flexible re-calibration of data-driven models is critical to ensure good forecasting performance. Then, a newly developed load-balancing control algorithm is presented for the coordinated action of distributed sewer assets using only current system states, rather than a full model of the sewer network. The algorithm incorporates flexibility to allow for the prioritization of multiple objectives to balance trade-offs between operational strategies and inform practical implementation.
Sara Troutman is an Environmental Engineering Ph.D. candidate at the University of Michigan, advised by Dr. Branko Kerkez and Dr. Nancy Love. She received M.S.E. degrees in Electrical and Computer Engineering and Civil Engineering (Intelligent Systems) from the University of Michigan, as well as B.S. degrees in Environmental Engineering and Mathematics from North Carolina State University. Her research broadly focuses on coupling environmental and water resources engineering, data science and machine learning, and systems analysis to tackle problems related to competing water quantity and quality objectives and the role of data in decision actions. She has received an NSF Graduate Research Fellowship and is a contributor to Open-Storm.org, an open-source repository of hardware, software, and algorithms for the sensing and control of water systems.