Bus travel time prediction with real-time traffic information
Introduction
As more vehicles are introduced into urban areas, traffic congestion has become a significant issue in many cities. One of the economic solutions to reduce pollution and traffic pressure is to develop Intelligent Public Transportation Systems (IPTS). Estimating travel times of public transportation is a critical component of IPTS since providing accurate real-time traveling information of public transits instead of pre-determined timetables can reduce passengers’ waiting time. It can also improve service quality of the system by instantaneously adjusting the departure schedule when unforeseen events occur.
Among different public transportation systems, the bus system is the most complex and densely distributed in modern cities, as it can use the existing road infrastructure and has less running costs. Hence, to improve the quality of public transportation service, we focus on the problem of how to accurately predict bus travel/arrival times in this paper. Providing real-time bus travel times has many applications including passengers’ trip planning (Bagchi and White, 2005), where the trips are scheduled by calculating the travel time of each destination, and bus route optimization (Yao et al., 2014), where system operators can design and adjust bus system networks according to the estimated travel times. Based on the above applications, many existing studies in predicting bus travel times have achieved varying degrees of success (Chien et al., 2002, Zhu et al., 2011). However, limitations still exist in previous studies.
First, bus data used in the field of predicting travel times, such as bus GPS data and smart card data, are relatively sparse when compared to other forms of location-based data, e.g., taxi and telephone trajectories, as a limited number of buses are in service (running) at the same time. It can result in bus data missing crucial real-time irregular traffic condition, such as accidents, that can affect the subsequent buses. An example is shown in Fig. 1. Due to an accident occurring between bus stops 1 and 2, the travel time of bus B01 may unexpectedly increase significantly. However, because of the sparsity of bus numbers, the former bus B02 cannot provide this traffic information for the next buses. As irregular traffic events have a significant influence on travel times (Hojati et al., 2016) and they are difficult to be captured by bus data alone, existing work fails to achieve accurate prediction results under complex traffic conditions.
Second, as bus routes are relatively long, previous models usually build different route construction methods to divide routes into segments and then to predict the travel times for each part. Two types of popular route construction methods divide the bus routes based on important road intersections and bus stop locations, which are called link-based and stop-based models, respectively. However, the bus dwelling times at stops are not considered independent of the transit times, where bus dwelling times represent the time periods the buses waiting at the stations, and the transit times represent the bus running times between every two stops. As shown in Fig. 1, bus dwelling times are significantly impacted by the number of passengers getting on and off, which is different from the transit times that are dependent on local traffic conditions. As shown in Fig. 2, affected by complex boarding patterns, bus dwelling times are more unstable and have different underlying patterns, which have a significant influence on total travel times. Hence there is a need to consider bus dwelling times independent of their transit times in order to increase the accuracy of the bus dwelling time estimation.
In order to address the limitations that we observed in the prevision works, as discussed above, a novel segment-based approach incorporating real-time traffic data is proposed in this paper. We divide the bus routes into transit and dwelling segments to predict total bus travel times, as shown in Fig. 3. There are two main technical challenges that we addressed in our approach. The first challenge is how to divide a bus route into transit and dwelling segments accurately? Existing bus data usually do not have information about the times the buses are “in” and “out” of the stops, and the online maps such as Google maps can only provide one single location for each stop. Further, the bus dwelling areas cannot be represented using a single point on the map, and it is challenging to capture this information using the raw bus GPS or online data. Therefore, we propose a methodology for automatic bus route segmentation that can rely only on the GPS data information. The second challenge is how to estimate bus transit and dwelling times accurately? As real-time traffic conditions are the main impact factors of transit times, while the number of passengers at stops is the primary factor that impacts the dwelling times, two independent travel patterns of transit and dwelling times are required to be discovered based on the different real-time traffic data. In summary, the main contributions of the proposed approach are described below:
- •
We propose a novel segment-based approach to predict bus travel times by learning bus transit and dwelling times from independent underlying patterns based on heterogeneous traffic data. It can increase prediction accuracy by reducing the uncertainty of dwelling times and dividing bus trips into regular and irregular traffic conditions. To the best of our knowledge, this is the first study to estimate bus travel times based on a combination of the bus and real-time taxi data. It reveals the impact of real-time traffic conditions on the bus travel time.
- •
We propose a method to build a segment-based bus route graph that can automatically and efficiently divide a route into dwelling and transit segments using bus GPS data only. This is the first work that only uses bus GPS to detect bus stop areas, which eliminates the process of labeling bus stops from online maps and avoids the influence of inaccurate labeled data to the prediction results. This helps to predict bus transit and dwelling time independently and accurately.
- •
We propose two models for both bus transit and dwelling times prediction. For bus transit times, an algorithm is proposed to use taxi speed data to recognize irregular bus trips and to predict them based on different weighted features. For bus dwelling times, a combination of passenger information and bus trajectories are used to help identify the dwelling time patterns of each stop. The experiment results show that the segment-based model can increase the prediction accuracy under both normal and abnormal traffic situations compared with the state-of-art methods.
Section snippets
Related work
Estimating the travel times of a route is an essential component for Intelligent Public Transport System (IPTS) (Yao et al., 2014), which has attracted many researchers and IPTS planners to address this problem. Various models have been proposed to predict bus travel, arrival time or speed, which can be summarized as shown in Fig. 4. We divide the existing approaches into three parts based on models, impact factors and route construction methods. A review and the comparison of existing studies
Overview
In this section, we first describe the key terminology and notation used in our approach and provide an overview of the proposed segment-based bus travel time prediction approach. Table 2 provides a summary of the symbols usded in this paper and their meanings.
Bus travel time prediction
In this section, we describe the details of our proposed segment-based bus travel time prediction approach. In our approach, the bus travel time of a trip is predicted by separating it into the bus transit time and dwelling time . and are predicted separately by two models with different traffic information. First, a segment-based bus route is built to divide bus routes into and . Second, a bus transit time model is built by adding real-time taxi information.
Evaluation
In this section, we evaluate the accuracy of our approach based on five types of experiments, which consist of the evaluation of basic bus transit time prediction model, abnormal bus trip detecting method evaluation, evaluation of bus transit time prediction model, evaluation of dwelling time prediction model and evaluation of bus travel time prediction approach.
Conclusion
In this paper, we investigate the problem of estimating public transport bus travel time with real-time traffic information. To solve such bus travel time prediction problem, a novel approach based on real-time data is presented in terms of the knowledge learned from a large number of vehicle trajectories and bus smart card data. In our method, a segment-based bus route graph is built to divide the bus route into different time slots and road segments. By separating the route into bus transit
References (31)
- et al.
The potential of public transport smart card data
Transport Policy
(2005) - et al.
The impact of real-time information on bus ridership in new york city
Transport. Res. Part C: Emerg. Technol.
(2015) - et al.
A prescription for transit arrival/departure prediction using automatic vehicle location data
Transport. Res. Part C: Emerg. Technol.
(2003) - et al.
Real-time bus route state forecasting using particle filter and mesoscopic modeling
Transport. Res. Part C: Emerg. Technol.
(2015) - et al.
Bus travel time prediction using a time-space discretization approach
Transport. Res. Part C: Emerg. Technol.
(2017) - et al.
Estimation of trip travel time distribution using a generalized markov chain approach
Transport. Res. Part C: Emerg. Technol.
(2017) - et al.
Analysis of bus travel time distributions for varying horizons and real-time applications
Transport. Res. Part C: Emerg. Technol.
(2018) - et al.
Where is my bus? Impact of mobile real-time information on the perceived and actual wait time of transit riders
Transport. Res. Part A: Policy Pract.
(2011) - et al.
Transit network design based on travel time reliability
Transport. Res. Part C: Emerg. Technol.
(2014) - et al.
Bus arrival time prediction at bus stop with multiple routes
Transport. Res. Part C: Emerg. Technol.
(2011)