Yiling Zhang - Research

Research

My research mainly focuses on developing application-motivated theory in data-driven optimization problems. In particular, I aim to address challenging decision-making problems in uncertain and adversarial environments, such as scheduling, routing, matching, and many others. They often involve integer decision variables and nonlinear constraints embedded in sequential decision-making processes. Overall, my research interests include

Methodology: Stochastic Programming, Integer Programming, Nonlinear Programming, and

Applications: Transportation and Logistics, Healthcare Operations, Power System Optimization

The following are some projects that I have been working on. Here are links to the reports in the media about some projects: Integration of routing and assortment optimization, Electrification of public transit systems.

1. Electric Vehicle-Sharing System Design and Operations

Driven by the growing environmental consciousness and rising vehicle ownership cost, the introduction and adoption of electric vehicle (EV) sharing draw increasing attention as a means to reduce greenhouse gas emission, traffic congestion, as well as transportation expenses. The large-scale vehicle-sharing problem involves complex planning and operational decisions, and multiple sources of uncertainties. We use a spatial-temporal network to model the system and related operations. We utilize stochastic programming techniques and decomposition algorithms to optimize system design and operations. Via numerical studies based on real-world data, we investigate the benefits of EV sharing for the service providers, users, socio-environment, and electrical grids.

2. Resource Planning and Operations

There are many stochastic service optimization problems with similar model structures, including healthcare operations, inventory control, cloud computing, etc. Our goal is to explore the problem structures to develop efficient and scalable computational schemes. For example, in the context of surgery planning in operating rooms (ORs), under the presence of uncertain surgery durations, the system operator needs to decide surgery block allocations and scheduling in each OR to provide a good service quality while maintaining an appropriate “end-of-the-day” guarantee for each OR not going overtime.

When perfect knowledge of the surgery durations’ probability distribution is not accessible, we consider risk-averse stochastic optimization models with distributionally robust optimization (DRO) techniques, which provide decisions that hedge against the unknown (ambiguous) probability distribution. The resulting formulations/approximations are convex (with binary decision variables). We derive cutting-plane methods and valid inequalities to improve computation.

3. Uncertain Load Control in Sustainable Power Systems

Increasing efforts have been made to utilize clean renewable energy sources (RESs), including wind and solar power generation. RESs are generally fluctuating and require additional reserves to manage real-time power imbalance. Load control provides such reserves, which, however, increases system uncertainty due to weather and load usage patterns. This project investigates data-driven and distribution-free approaches to managing the time-varying, correlated, and complex uncertainty associated with load control in power system scheduling problems with fluctuating RESs. We focus on investigating, characterizing, and managing this uncertainty. Moreover, we quantify the tradeoff between the uncertainty and profitability, and assess the impact of uncertainty on environmental sustainability.