Relocation strategies for e-scooter system optimization

In the last few years, the concept of micromobility has gained a lot of ground, thanks to the introduction of new transportation modes, like e-scooters. These new, lightweight, electric vehicles are cheap, easy and fun to ride, and they are increasingly becoming a reasonable solution for first- and last-mile trips. New companies started spreading e-scooters throughout different cities\\around the world, with the aim of offering a new, dockless, e-scooter sharing service. These companies are growing fast and they have already been able to attract several hundred million dollars of investments. E-scooters - and micromobility in general - have the capability to reduce pollution and traffic congestion, but researches are questioning if they are really positively contributing to solve these issues. It seems that, moving e-scooters through the city with other motorized vehicles for charging, deployment and relocation, is a major component of their entire life-cycle emissions. Moreover, an even bigger contribution to these emissions is given by manufacturing. Intensive use and vandalism cause a shorter vehicle lifetime and, consequently, an higher e-scooter production demand, thus increasing emissions associated with manufacturing. In addition, traffic congestion is reduced, but sidewalks are becoming more and more populated with wrongly parked e-scooters, and they are becoming the preferred path for e-scooter riders, when bike lanes are not present. It is therefore useful to keep the number of deployable e-scooters low, and to try to maximize their utilization, repositioning them in a reasonable way, possibly also reducing the total amount of transportation distance. In this thesis, we analyze different relocation algorithms, in order to understand, first of all, if it is useful and profitable to relocate. We show how different relocation strategies affect the system, both in terms of performance and costs. For this purpose, we adopt and extend an existing data-driven, discrete-event simulator for Free-Floating Vehicle Sharing Systems (FFVSS). We introduce new datasets of e-scooter trips, cleaning and preparing them to be utilized with the simulator, and reaching a total amount of 7 North-American cities. We extract a demand model from data and we use it as input for the simulator. We introduce new reactive and proactive algorithms, with the aim of solving the relocation problem with a greedy approach. We test each algorithm under different scenarios and we make cost and revenues assumptions to study the profitability of the system. Results show that, even with a greedy solution, it is possible to relocate in a useful way. However, a trade off is needed between total number of deployed e-scooters and number of workers in charge of relocation, in order to make the system profitable.