HIL Seminars
Direct Policy Search for Multi-Objective Control of Socio-Ecological and Hydrological Systems
Julianne D. Quinn
Cornell University, Ithaca, NY
DEIB - Seminar Room
March 7th, 2017
3.00 pm
Contacts:
Andrea Castelletti
Matteo Giuliani
Research Line:
Planning and management of environmental systems
Julianne D. Quinn
Cornell University, Ithaca, NY
DEIB - Seminar Room
March 7th, 2017
3.00 pm
Contacts:
Andrea Castelletti
Matteo Giuliani
Research Line:
Planning and management of environmental systems
Sommario
On March 7th, 2017 at 3.00 pm, "Direct Policy Search for Multi-Objective Control of Socio-Ecological and Hydrological Systems" seminar will take place at DEIB Seminar Room, as new appointment of HIL Seminars.
Designing management strategies for socio-ecological and hydrological systems requires balancing competing societal and environmental objectives. This challenge is compounded by uncertainties in the appropriate system model structure, parameters, inputs, and even objectives. Using two different system models, we show that direct policy search (DPS) is a flexible and computationally efficient method capable of handling these challenges. The first system is a managed lake system in which a town must balance the economic benefits it derives from discharging phosphorous into the lake against the environmental risks of tipping the lake into a permanently polluted state. In this system, DPS is able to quickly find effective pollution control policies that are robust to parametric uncertainties in the lake’s phosphorous dynamics. The second system is a multi-reservoir system in the Red River Basin in Vietnam that must be operated to balance competing needs for hydropower production, agricultural water supply and flood protection to the capital city of Hanoi. In this system, we use DPS to optimize operating policies for the reservoirs under different mathematical formulations of objectives in the optimization problem. We find that some formulations suffer severe unintended consequences that would go unnoticed without a flexible optimization framework capable of testing such rival problem framings.
Designing management strategies for socio-ecological and hydrological systems requires balancing competing societal and environmental objectives. This challenge is compounded by uncertainties in the appropriate system model structure, parameters, inputs, and even objectives. Using two different system models, we show that direct policy search (DPS) is a flexible and computationally efficient method capable of handling these challenges. The first system is a managed lake system in which a town must balance the economic benefits it derives from discharging phosphorous into the lake against the environmental risks of tipping the lake into a permanently polluted state. In this system, DPS is able to quickly find effective pollution control policies that are robust to parametric uncertainties in the lake’s phosphorous dynamics. The second system is a multi-reservoir system in the Red River Basin in Vietnam that must be operated to balance competing needs for hydropower production, agricultural water supply and flood protection to the capital city of Hanoi. In this system, we use DPS to optimize operating policies for the reservoirs under different mathematical formulations of objectives in the optimization problem. We find that some formulations suffer severe unintended consequences that would go unnoticed without a flexible optimization framework capable of testing such rival problem framings.
Biografia
Julie Quinn is a PhD student at Cornell University. Her research centers on multi-objective management of water resources systems under uncertainty.