Combining Profile Similarity and Kalman Filter for Real-World Applicable Short-Term Bus Travel Time Prediction

The emergence of intelligent transportation systems and the availability of detailed past trips and real-time data enabled more accurate travel time prediction algorithms. Passengers of public transportation highly value reliable service and accurate travel information. Unexpected delays while traveling impact traveler satisfaction enormously. We present a novel approach for a travel time prediction model for public transit in an urban setting. For this, we extended the existing profile similarity model based on k-medoids clustering and augmented it for an urban regions. The algorithm integrates real-time data from preceding vehicles traveling on the same links with a Kalman filter. Hence, the developed model uses the information available in modern ITS by combining historic travel time profiles with real-time data. While the algorithm focuses on short-term predictions, we have also evaluated it for long-term forecasts. Our results show that the proposed model outperforms benchmark models regarding certain criteria and can provide accurate travel times with comparably low-volume data input. Furthermore, the results indicate that the combination of a Kalman filter with a k-medoids approach improves the quality of the predictions.

I. Introduction

OVER the last decades, intelligent transportation systems (ITS) for public transport have evolved into distributed, highly complex, and versatile systems. For passengers, the predominant task of an ITS is providing accurate information on public transportation to public displays or personal smartphones in real-time. Today, this information includes the planned arrival times as well as the predicted arrival times of vehicles. An ITS also assists public transit operators by permanently monitoring the fleet, allows for control of the fleet, and stores all operational information for later use. Therefore, with increasing ITS availability, operational data's overall availability, including past trips of vehicles, also rises. This increase of available data enables more sophisticated travel time prediction algorithms.