Recommendation of feeder bus routes using neural network embedding-based optimization
mode - bus, mode - subway/metro, mode - taxi, place - urban, technology - geographic information systems, technology - intelligent transport systems, planning - route design, planning - methods
Transportation, Road2vec, K-means clustering, Integer programming, Feeder bus routes
Despite the existence of vast bus and subway networks, the demand for taxis in the morning commute hours is substantial in metropolitan areas. This kind of morning traffic can be resolved by means of feeder buses connecting residential areas with popular transit points. To assist the design of feeder bus routes, this study proposes an optimization approach based on road2vec, which is applied to real-time taxi GPS data. Road2vec is a neural network-based embedding methodology that extracts road name vectors considering the movement patterns of vehicles. Subsequently, the k-means clustering analysis is applied to those vectors to identify the major taxi transit clusters during the commute hours. For each cluster, we suggest a feeder bus route that can best reflect the taxi trajectory patterns. To find intermediate stops between the departure and origin of a feeder bus route, we solve an integer programming to maximize the cosine similarity between the origin road vector and the departure road vector subtracted by the road vectors of intermediate stops. The suggested routes based on our method differ from the existing routes in that they have a tendency to pass through residential areas, transit stations, and schools. In addition, the result suggests that the model developed in this study finds bus routes that could be suitable for feeder buses by accommodating areas where the demand for taxis is high in the morning. Our road2vec approach is expected to contribute to a reduction in traffic during rush hours.
Permission to publish the abstract has been given by Elsevier, copyright remains with them.
Park, C., Lee, J., & Sohn, S.Y. (2019). Recommendation of feeder bus routes using neural network embedding-based optimization. Transportation Research Part A: Policy and Practice, Vol. 126, pp. 329-341.