CAREER: Incorporating Decision-Dependent Uncertainty via Distributionally Robust Optimization: Models, Solution Approaches, and Applications

点击翻译按钮获取中文摘要

A wide range of decision making problems in the service industries (e.g., power, healthcare, and transportation) involve uncertain parameters, which partially depend on the decision itself and hence are decision-dependent or endogenous. Incorporating such dependency in the decision-making process is significant, because otherwise these uncertain parameters may unexpectedly undermine the system performance. Furthermore, such dependency provides an opportunity of proactively maneuvering the uncertainty to reinforce the decision making. This CAREER project will evaluate the potential benefit of incorporating decision-dependent uncertainty and investigate new optimization approaches to maneuvering such uncertainty. If successfully implemented in the service industries, the research findings of this project will improve the sustainability of the society and the well-being of the people. In addition, this project will develop new educational materials including a power system simulation game and a graduate-level course on stochastic and robust optimization, which help inspire the next generation of engineers.As compared to the exogenous uncertainty that is independent of the decision, endogenous uncertainty has received much less attention in the applications. The goal of this CAREER project is to holistically quantify, model, and maneuver the endogenous uncertainty via distributionally robust optimization. The intellectual significance of this research includes (a) providing provable out-of-sample performance guarantee for making decisions under endogenous uncertainty and (b) mitigating the computational challenges of incorporating endogenous uncertainty in optimization problems. Specifically, the scope of the research includes (1) quantifying the value of considering endogenous uncertainty when it actually arises, (2) modeling the endogenous uncertainty by a family of decision-dependent probability distributions, (3) proactively maneuvering the endogenous uncertainty to improve system performance, and (4) applying the resulting methodology in applications that involve endogenous uncertainty, e.g., do-not-exceed limits, demand response programs, and off-shore oil drilling.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.