My current research employs stochastic dynamic programs and Markov decision processes to answer questions regarding electric vehicle routing and access to charging infrastructure.

Until a working paper is ready (expected later this year), you can learn more about the project by expanding the panel below:

The EVRP-MRUA: Heuristic policies and lower bounds

Motivated by environmental concerns and regulations, electric vehicles (EVs) are becoming more popular in supply chain distribution functions (e.g., La Poste, UPS, Coca Cola). However, EVs pose operational challenges to which their conventional petroleum-based counterparts are immune. For instance, EVs' driving ranges are often only 25 percent that of conventional petroleum-based vehicles' (CVs), charging infrastructure is still relatively sparse compared to the network of refueling stations for CVs, and the time required to charge an EV can range from 30 minutes to 12 hours depending on charging technology - orders of magnitude longer than the time needed to refuel a CV.

There are two general approaches to overcoming these operational challenges. The first is a simple approach in which routes are restricted to the vehicle's autonomy. That is, the EV is routed back to the depot when its battery nears depletion so it may charge overnight in preparation for the subsequent day's deliveries. In the second approach, the EV is allowed to perform mid-route recharging by taking advantage of charging infrastructure in the field.

Past research has shown that the second approach offers cost savings, because mid-route recharging allows for a decrease in the total distance traveled and an increase in the capacity of a single EV, thereby reducing the number of vehicles and drivers needed. However, these studies that have considered mid-route recharging make the assumption that the charging stations (CSs) are always available to the EV when it arrives to charge. In reality, this is often not the case. Because charging station infrastructure is limited and EVs require significant time to charge, charging stations will often be unavailable when an EV arrives, and the EV may be forced to queue. This discrepancy between modeling assumptions and reality has thus far prohibited logistics companies from implementing mid-route recharging, despite the suggested cost savings.

Our research aims to reduce this discrepancy by more realistically modeling both the uncertainty in availability and the queuing process at public charging infrastructure. We model the EV Routing Problem with Mid-route Recharging and Uncertain Availability (EVRP-MRUA) as a Markov decision process and implement a stochastic dynamic programming (SDP) solution.

In addition to developing a handful of heuristic policies to solve the SDP, we also employ an information relaxation to determine the optimal solution with perfect information. This gives us a lower bound and therefore a measure of goodness for our policies. We do this by formulating a mixed integer linear program that is equivalent to the SDP.

Further, we impose information penalties to tighten the lower bound. Current results indicate that our heuristic policies are performing well - within 5% of the optimal policy.

Our work aims to enable logistics companies to take advantage of the increases in capacity offered by mid-route recharging, thus extending the utility of EVs as delivery vehicles.

Slides from my presentation at the first triennial INFORMS TSL Conference can be found here.

Previously, I used multi-objective math programs to solve problems regarding natural resource management under alternative climate change scenarios. These problems served as a platform on which to study the conflict among objectives in multi-objective optimization.

You can learn more about this research here.