Dynamic Rebalancing Optimization for Bike-Sharing System Using Priority-Based MOEA/D Algorithm

Abstract

As an indispensable part of public transportation systems, the bike-sharing system (BSS) can improve road resource utilization and alleviate traffic congestion, significantly improving urban mobility. The disproportion between the demand and supply creates a giant gap for maintaining the smooth functioning of the system. To address the issue, this paper proposes a dynamic optimization rebalancing model for docked bike-sharing systems, which aims at minimizing the operation cost of rebalancing while maximizing the user satisfaction. The rebalancing demand is evaluated using both historical and predicted data so that a second time service for each station could be avoided within a rebalancing horizon. A time-window based satisfaction modeling is put forward to evaluate user satisfaction. Multiobjective evolutionary algorithm based on decomposition (MOEA/D) under the rolling horizon strategy is adopted to solve the model. To improve the algorithmic performance, local search based on station priority is applied. Numerical experiments using real-world data were implemented to demonstrate the proposed model and the advantage of the improved algorithm. As the results indicate, the proposed algorithm outperforms the nondominated sorting genetic algorithm II (NSGA-II) and MOEA/D without prioritybased local search. Solutions with higher satisfaction profit can be discovered with the help of a local search heuristic based on station priority. The experiment also revealed that a slight increase of 7.02% (¥29.2) in the rebalancing cost could yield a significant growth of 1233.7% (¥288.7) in satisfaction profit.

Intelligent Shared Mobility Systems: A Survey on Whole System Design Requirements, Challenges and Future Direction

“…The number of completed trips can measure the evaluation of re-balancing [70] or the ratio of failed requests [28] after re-balancing. VOLUME 10, 2022 Centralised re-balancing solutions use single or multiobjective optimisation techniques such as mixed-integer linear programming [1], [28], [71]- [74], heuristic search algorithms [5], [15], [16], evolutionary computation [75], stochastic processes [6] and deep reinforcement learning [69] to re-balance supply and demand within the whole road network. Additionally, to minimise the staff operation cost, an incentive is used to encourage drivers to pick up vehicles from a particular station that has a higher supply [70], [71].…”

“…Re-balancing the bike supply over multiple locations to meet the dynamic users demand is a well-studied challenge [16], [72], [75], [94]- [97]. The core of re-balancing is often solved by using the mixed integer programming approach [5], [16], [72], [98].…”

“…Offering an efficient re-balancing solution often implies improving the time complexity of the process in order to speed up the system's operations [7], [13], [16], [28], [30], [62], [72], [91], reducing costs/improving profits [3], [15], [21], [26], [75], [95], [98] or converging to an acceptable pricing strategy [5], [71], [96], [103]. The time complexity is improved by problem parallelisation or spatial/temporal simplifications.…”

“…In the context of centralised systems, it can be explained by the fact that it only affects drivers' decisions to re-balance a vehicle, when it is in the economic interests of the driver to re-balance. The introduction of parameters that include the drivers' preferences [75], [137] may help to reach a more acceptable re-balancing solution. Considering a decentralised system with autonomous agents making decisions [21], [24], [25], the autonomy is automatically implemented.…”

DeepBike: A Deep Reinforcement Learning Based Model for Large-scale Online Bike Share Rebalancing

“…Their solutions rely on complete data of the entire planning horizon, but in dynamic rebalancing, customer demands for the upcoming planning horizon are unknown to the decision-makers. In recent years, to generate online rebalancing solutions, heuristic algorithms and artificial intelligence-based approaches (e.g., deep learning and reinforcement learning) have been employed (Hu et al, 2021;Li et al, 2018;Shui & Szeto, 2018). However, these algorithms were only designed for and evaluated on small-scale BSSs (typically less than 300 bike stations) or they divided the study region into small clusters and dispatched only one rebalancing vehicle in each cluster to simplify the problem (Vallez et al, 2021).…”