Detecting congestion in bus terminals using historical positioning data

Efficient operation of bus terminals is crucial to the performance of large public transportation systems, particularly in urban environments where space constraints limit physical expansion. This thesis investigates methods for detecting and analyzing congestion in bus terminals using Automatic Vehicle Location (AVL) positioning data. The study utilizes historical real-time and static transit data in the General Transit Feed Specification (GTFS) format, combined with video footage from a drone for validation. A methodology is developed to detect stop occurrences, classify them according to their cause (e.g. at a berth, at traffic lights), validate these classifications using the video observations, and isolate the possible cases of congestion between buses. Data cleaning techniques and algorithmic adjustments are presented to improve detection accuracy. The main result of this approach is a table containing the occurrences of vehicles stopping within a determined time frame, with the detected possible stopping causes as an attribute for each occurrence. A numerical analysis is also conducted to assess berth occupancy and travel times within the terminal. The methodology is applied to the case of a typical Swedish bus terminal, Linköping Central. The proposed method involves recording the vehicle stop occurrences and the possible reasons for each stop. The algorithm detects the berth the vehicle is stopping at, if any, as well as the berths it is supposed to stop at according to its schedule. It also detects other possible reasons, such as at the traffic light at the exit of the terminal, at a pedestrian crossing or because of another stopped vehicle ahead of them. Validation of the results using the video observations is fully integrated within this process. Applying the methodology to the case of Linköping Central allowed for its evaluation, as well as several findings on the cases of congestion from the dataset to be drawn. The results from the berth detection algorithm were deemed as satisfying, as well as the detection of “regular” stops (in this case, traffic lights at the exit), while the algorithms for the detection of other stopped vehicles ahead and pedestrian crossings led to poorer results. All 6 cases of congestion found in the dataset happened near a different berth, but all of them occurred either because of a bus badly parked at its berth or another vehicle attempting to stop at an already-occupied berth. However, the dataset was restricted by the availability of drone footage, which might have limited too much the occurrences of congestion. Lastly, the numerical analysis led to the observation that vehicle distribution among berths was not optimal, as well as that some particular berths were potentially overloaded. The study proposes a generalizable framework, offering potential for broader application across different terminals and transit networks. As it is also a pre-study for a larger project, several suggestions are made for further work on the topic, including the expansion of the hereby presented approach, some other approaches such as microsimulation, and the possibility of taking into account other factors such as pedestrian movement and the needs of the stakeholders.