E-Scooters in Tucson, AZ: Modeling placement, charging, and rebalancing

"The decision making process for e-scooter companies is complex. One of the first questions is where to locate the scooters -- In the transportation network, where do e-scooters need to be placed to meet demand? The second question is how to distribute them. It gets more complicated when you introduce different electric charging methods, so that some scooters are being collected by paid contractors and others are being charged by customers, through incentives," Cheng said.

As the researchers see it, the main benefits of shared mobility are threefold:

1. It's an affordable first/last mile option for people who cannot afford to own a vehicle;
2. It's environmentally friendly, using fewer resources more efficiently;
3. It offers a novel solution to the problem of traffic congestion.

The model they created can provide decision makers with a robust solution that enables low cost and high service quality.

What Does the Model Do?

Cheng, along with Dr. Xiaofeng Li, Dr. Yao-Jan Wu, and doctoral student Abolhassan Mohammadi Fathabad, created a two-stage stochastic programming model, for the planning and operation of large shared mobility networks in the presence of demand uncertainty. The model, using a random probability pattern, can answer several questions that arise at the company and policy-maker levels:

• In the face of uncertain demand, how many e-scooters should be placed in the network and in which locations?
• How many charging facilities are needed, where and at what capacities?
• How should the network be evaluated and adjusted daily, to continue to meet demand while minimizing costs?

The model, developed using data from the City of Tucson, allows for optimal placement, charging, and rebalancing of unused (idle) scooters to meet demand in the most efficient manner possible. Successful operation of a shared mobility system requires both careful planning and strategies that minimize operational costs for the company while also increasing the customer satisfaction rate (i.e., an e-scooter is available when it is wanted). From a longer-term planning perspective, it is necessary to consider e-scooter demand patterns on a seasonal basis and plan accordingly for long-term economic benefits. The final report offers a detailed walk-through of the computational model.

Possibilities for Future Research

By predicting scooter rentals, relocations, idle periods, and charging times, the model offers e-scooter management companies a ready-made decision-making system to to effectively design and operate shared e-scooter systems, and thus help to ensure system reliability and cost-effectiveness.

In the future, this research could be extended in several directions. First, to increase the robustness in uncertainty modeling, researchers could develop a distributionally robust optimization framework (a modeling framework for decision-making under uncertainty). Another possible direction is to develop new approaches to solve the e-scooter planning problem when more variables are introduced that would require the model to be capable of solving more complex problems.