Akbari-Moghaddam, Maryam (McMaster University), Kelly, Stephen (Mcmaster University), Down, Douglas (McMaster University)

Demand Forecasting and Rebalancing in Shared Bike Systems Using Deep Learning and Evolutionary Computation

Scheduled for presentation during the Regular Session "Transport planning" (FrAT9), Friday, September 27, 2024, 11:30−11:50, Salon 17

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 7, 2024

Abstract

Shared bikes offer an eco-friendly alternative to conventional public transport and can reduce traffic congestion. However, imbalances in bike availability at stations necessitate effective rebalancing strategies to prevent shortages or surpluses. Previous studies on shared bike rebalancing have mainly concentrated on station demand forecasting or route optimization. However, focusing only on demand forecasting does not effectively manage bike quantities at stations, and route optimization alone fails to address real-time demand fluctuations. This paper introduces a hybrid solution combining deep learning and evolutionary computing to tackle both demand forecasting and route optimization for rebalancing with multiple capacitated trucks. Station demand forecasting is modeled as a time series forecasting problem, and route optimization for bike rebalancing, guided by these forecasts, is addressed as a Capacitated Vehicle Routing Problem with Pickup and Delivery (CVRPPD). Deep learning is used to predict short-term demand at each station, which then informs the rebalancing strategy. Our objective is to optimize rebalancing routes to minimize both unmet station demand and carbon emissions from trucks. This involves selecting which stations each truck should visit. We use a Genetic Algorithm (GA) to identify the most efficient rebalancing routes.