Designing Flexible-Bus System with Ad-Hoc Service Using Travel-Demand Clustering

Providing direct and affordable transit services for travelers is the goal of the evolving flexible-bus (FB) system. In this study, we design an FB system with an ad-hoc service, to supplement traditional public transit and provide a better FB service. We first build up a mathematical model to optimize bus-stop sites, routes, and schedules, where the unmet travel demand is served by an ad-hoc service with relatively high cost. Then, we cluster travel demand spatially and temporarily, using the ST-DBSCAN algorithm. We use the simulated-annealing algorithm, which has better convergence and diversity than other heuristic algorithms, to solve the suggested model in large-scale networks. To demonstrate the effectiveness of the proposed model, we run experiments on a small network and a large real-world network of Shenzhen airport, which shows that the FB system with ad-hoc service can reduce overall cost and improve social welfare, compared to taxies and FB only. In addition, it provides affordable transit services with shorter walking distances and lower waiting times, which can be deployed in airports or high-speed railway stations with massive, irregular travel demands.