Vehicle delay series forecast based on trajectories of GPS tracked cabs

The vehicle delay serves as effective criteria for evaluating and optimizing the level of service of intersections in traffic. Various methods based on analytical models, simulations, or sensors like GPS have been proposed for estimating vehicle delay. However, the absence of large scale fine-grained GPS data limits its wide application. Taking advantages of available coarse-grained trajectories of GPS tracked cabs, this paper proposes alternative definition of vehicle delay and its time series in pursuit of capturing long-range characteristics of traffic. The forecast performance of several vehicle delay series analysis algorithms are compared. A new Robust TBATS algorithm is proposed to predict vehicle delay series with outliers. 8-month trajectories of data in the city of Shenzhen, P.R. China are utilized for quantified comparison. The test verifies the value of vehicle delay series. The proposed Robust TBATS algorithm is capable of automatically forecast and its forecast performance outperforms other baselines significantly. Furthermore, this paper demonstrates one of potential applications of vehicle delay series forecast in route planning.