Optimal Rebalancing Strategy for Shared e-Scooter Using Genetic Algorithm

Abstract

Shared e-scooters are provided as a free-floating service that can be freely rented and returned within the service area. Although this has a positive effect in terms of convenience for users of shared e-scooters, it is creating new urban problems, such as undermining the aesthetics of the city and obstructing the passage of pedestrians. Therefore, this study developed an optimal rebalancing algorithm to mitigate these problems and proposed an efficient operation plan. Complete relocation was performed to match the demand and supply for an efficient operation by reducing the unnecessary oversupply of shared e-scooters. The optimal rebalancing algorithm that reflects the attributes of e-scooters was developed through genetic algorithms and subsequently applied to actually used cases. The results indicate that when 20% of the potential demand was considered, an optimal solution could be derived with two relocation vehicles; however, when the potential demand was not considered, three relocation vehicles were required. Therefore, it is anticipated that the results of this study can serve as basic data for solving various urban problems caused by the recent rapid increase in the use of shared e-scooters.

Optimizing Shared E-Scooter Operations Under Demand Uncertainty: A Framework Integrating Machine Learning and Optimization Techniques

“…Shui and Szeto [45] reviewed studies and found that they have focused on different characteristics of shared-bike rebalancing, such as objective functions (e.g., for distance, cost, and emissions), constraints (e.g., budget, service time, and inventory), optimization algorithms (exact or heuristic algorithms), deterministic or stochastic problems, and static or dynamic problems. Several other studies also summarized these challenges of rebalancing problems, including methodology [46], objectives [47,48], problem size [49], number of rebalancing vehicles [50,51], damaged bikes [52], equilibrium of station [53], and multi-step matching [54]. The present study focused on problems of rebalancing shared bikes under demand uncertainty and with predicted demand.…”

“…Various heuristic algorithms have also been developed or applied to address these NP-hard optimization problems. These include (hybrid) genetic algorithm [48,71], ant colony optimization with CP (ACO-CP) [77], discrete-continuous hybrid model [38], CA [35], extended particle swarm optimization [52,53], greedygenetic heuristic [46], tabu search [49], neighborhood search [50], and neighborhood search-variable neighborhood descent [51]. Other algorithms can be found tabulated in the references [46][47][48]50].…”