Development of an AI-powered shareable bike rebalancing system

• Main Author: Tu, Anqi
• Other Authors: Yu Han
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Software::Software engineering; Engineering::Mathematics and analysis::Simulations
• Online Access: https://hdl.handle.net/10356/138644
• Institution: Nanyang Technological University

Bike sharing systems with docking stations are widely deployed in many major cities, bringing convenience to citizens and promoting eco-friendly lifestyles. However, they are facing a common problem - the congestion or deficiency of bikes in docking stations due to fluctuation in bike usage. Inefciency in re-distributing bikes among docking stations is challenging for system operators. One approach to address such inefficiency is to rebalance bikes among docking stations with trucks. To allow researchers to study the efficiency of different rebalancing strategies under different conditions, this project aims to develop a simulation testbed. This paper presents Rebalancer, an AI-powered Shareable Bike Rebalancing System, which is capable of loading real-world bike sharing system datasets to simulate collective usage behaviours. It is integrated with well designed models for spatial-temporal traffic prediction, and is incorporated with a spatial-temporal rebalancing algorithm as a default approach for users to adjust and extend. It allows the user to interactively simulate and evaluate the dynamic rebalancing operations of shareable bikes, providing visualization of the AI decisions, the movement of bikes, and the trucks used to re-distributing bikes. Compared to other purely machine learning-based approaches, this testbed allows system operators to incorporate their preferences and business constraints into the rebalancing operations to be visualized and evaluated under realistic conditions. Through simulation with London bike sharing system’s dataset retrieved from the Transport for London website, Rebalancer demonstrates effectiveness of the spatial-temporal rebalancing algorithm in reducing the demand and supply gap of bikes in docking stations. Experiments are also conducted to study the performance of different models in predicting traffic for each station. The results show LSTM yields the best performance, with lowest root-mean-square error and highest stability.