Development of an Agent Based Simulation Framework for Public Bus Operations to Improve Arrival Reliability



Development of an Agent Based Simulation Framework for Public Bus Operations to Improve Arrival Reliability

Authors

Muhamad Azfar Ramli, Institute of High Performance Computing, A*STAR, Vasundhara Jayaraman, Institute of High Performance Computing, A*STAR, Kwek Hyen Chee, SMRT Buses Ltd

Description

A framework to simulate bus transit operations is presented, combining models for dispatch deviation, transit speed and dwell time. We apply the simulation by identifying causes of headway instability and study factors affecting arrival reliability.

Abstract

We present a framework for modeling bus transit operations using a discrete event, agent-based simulation combining: i) a dispatch model ii) a transit speed model ii) and a dwell time model adapted from literature. Model parameters are derived from analyses of a 3 month long historical data collected from a major trunk service route obtained from a local transport operator. The data consists primarily of three components: 1) automated vehicle location (AVL) data that provides stop-to-stop vehicle timings and 2) automated passenger farecard collection (APC) data and 3) bus schedule data. We discuss how the integration of these three datasets improves the overall accuracy of the datasets and allows for more accurate modelling of operation-specific constraints as compared to previous related work in this area.

Past simulation frameworks typically utilize traffic or micro-type models in order to model the bus transit speed. Given limited access to more granular data, we show that statistical fitting can effectively model the impact of traffic fluctuations on bus transit speed in different road segments. We also show that transit speed distributions can be simulated fairly accurately using many different generic skewed distributions.

We also adopt a dwell time model that matches the context of our local buses; taking into consideration not only the number of passengers boarding and alighting but also the crowdedness of the buses as a possible cause of additional dwell time delay. The model is recalibrated for our purposes and the results closely match previously obtained values for the average commuter boarding and alighting rates.

By including actual bus schedules into the simulation framework, we are also able to study the deviations found between the schedule against actual arrivals and departures of buses and learn that these errors are not only caused by random human error but also due to late arrivals at the interchange. These deviations are therefore also affected by the constraint on fleet numbers available for dispatch.

Through the agent-based simulation framework, we perform scenario based modeling such as changes in commuting demand and the effect of schedule changes on the reliability of arrivals. This procedure can potentially help transport operators to identify the baseline performance of a given bus service under normal operating traffic conditions. It also allows operators to test out optimal strategies prior to testing using actual bus operations so as to help them maintain operational standards under increasingly adverse circumstances.

Publisher

Association for European Transport