BusWTE: Realtime Bus Waiting Time Estimation of GPS Missing via Multi-Task Learning

Abstract

Realtime bus waiting time is of great importance to the intelligent public transportation system and is beneficial for improving user satisfaction by online map services. While there are limited realtime bus waiting time services in a city, because of the expensive cost of GPS sensor deployment and realtime service operation. To address the above problem, we propose a novel end-to-end multi-task framework named BusWTE, which estimates bus waiting time for those bus routes without GPS sensors deployed. BusWTE utilizes a variety of urban datasets, including historical bus trip data reported by a limited number of GPS equipped buses, road network data, traffic condition data, and mobility data. Specifically, we firstly use a classical BiLSTM architecture to encode the sequence of bus route related features, and employ two fully-connected layers to embed the stop related features and temporal features, respectively. Then a temporal attention mechanism is proposed to capture the dynamic correlation between the route features and temporal features. Furthermore, we employ multi-task learning to estimate the bus waiting time and the bus interval simultaneously, which highly improves the model performance. Finally, extensive experiments conducted on two large-scale real-world datasets demonstrate the effectiveness of BusWTE. In addition, BusWTE has been deployed on Baidu Map app, servicing over twenty major cities in China.