A data-driven agent-based model of congestion and scaling dynamics of rapid transit systems

doi:10.1016/j.jocs.2015.03.006

Journal of Computational Science

Volume 10, September 2015, Pages 338-350

https://doi.org/10.1016/j.jocs.2015.03.006 Get rights and content

Highlights

•
Develop a full-scale data-driven agent-based model (ABM) of rapid train system (RTS).
•
Model integrates a route choice model and results are validated using smart card data.
•
Analyze congestion and scaling dynamics of RTS using the developed ABM.
•
Propose a procedure that captures the penultimate station effect in RTS.

Abstract

Investigating congestion in train rapid transit systems (RTS) in today's urban cities is a challenge compounded by limited data availability and difficulties in model validation. Here, we integrate information from travel smart card data, a mathematical model of route choice, and a full-scale agent-based model of the Singapore RTS to provide a more comprehensive understanding of the congestion dynamics than can be obtained through analytical modelling alone. Our model is empirically validated, and allows for close inspection of congestion and scaling dynamics. By adjusting our model, we can estimate the effective capacity of the RTS trains as well as replicate the penultimate station effect, where commuters travel backwards to the preceding station to catch a seat, sacrificing time for comfort. Using current data, the crowdedness in all 121 stations appears to be distributed log-normally. We find that increasing the current population (2 million) beyond a factor of approximately 10% leads to an exponential deterioration in service quality. We also show that incentivizing commuters to avoid the most congested hours can bring modest improvements to the service quality. Finally, our model can be used to generate simulated data for statistical analysis when such data are not empirically available, as is often the case.

Introduction

To tackle rising population density in urban cities, transportation planners often construct train rapid transit systems (RTS) as a first step. Yet continued population growth forces the RTS to evolve towards increased complexity with more train lines added to satisfy demand. With the increased complexity, planners are confronted with the difficulty of predicting commuter ridership, route choices, and also the various outcomes of the system during disruptions. Moreover, increased station and train crowdedness in RTS lead to congestion, commuter discomfort, trip delays, and lowered overall service quality standards. It is therefore imperative that modern transportation models be capable of investigating not just the issues of efficient, robust and scalable transportation, but also of commuter comfort and satisfaction.

The introduction of smart card ticketing in RTS has serendipitously enabled large-scale data analytics into commuter travel behaviour [1], [17]. Analytical and regression models have been developed to estimate commuters’ spatio-temporal density [20], identification of boarded trains [10], travel patterns [4], and transit use variability [14]. Yet, it is noted that the information captured by smart cards has limitations [17]; for example, most datasets do not contain routing information as they capture information only at the entry and egress points of journeys.

In contrast to analytical and regression models, agent-based models (ABM) strive to model each individual agent in a manner most natural to the system at hand [3]. Essentially, an agent is autonomous and formulates decisions and interacts with other agents directly. By directly replicating the mechanics of the system, an ABM permits the observation of emergent phenomena that arise from the interactions of the agents with each other [3] – provided the mechanics are correctly characterized and the model is well-calibrated.

ABM has seen recent success in modelling large-scale transportation [7], [15], [21]. However, there are not many studies which incorporate smart card data to drive RTS demand for better calibration. In our previous work [11], we had leveraged upon anonymized travel smart card transactional data to synthesize travel demand for a smaller-scale agent-based model of the Singapore transit system involving only one of the operational train lines, and achieved a very close match between the simulated and empirical travel duration distributions. In that work, we also investigated the dynamics of the smaller-scale system with regard to population growth.

Here, we extend our previous work [11] by: (1) expanding the model to cover all seven operational lines; (2) adding a route-choice mechanism inferred statistically from empirical travel duration distributions [13]; (3) incorporating station-specific walk-times; (4) investigating dynamics that were not directly measurable in our dataset, such as station crowdedness; (5) estimating the effective train capacity; (6) modelling the penultimate station effect; and (7) running further population scaling scenarios. We validate our model by ensuring the travel duration distributions generated from our simulations are well-calibrated to the empirical dataset. This would lend strength to any inferences derived from our scenarios. Apart from these goals, ultimately, we strive to construct a simulation platform that can be used to evaluate the efficacy of proposed strategies in tackling current and future urban transportation issues.

Our experiments in this work are focused on the Singapore rail transit system, which began operations in 1987 and is now one of the busiest RTS in the world. Despite the focus, our approach can be applied to other rail transit systems in the world, as we do not utilize any Singapore-specific mechanics or assumptions in our model.

Section snippets

Data

Our model is dependent on data for the following purposes: (1) to construct the transit infrastructure, (2) to instantiate the commuter agents corresponding to the actual travel demand, (3) to calibrate the travel time components of the network, and (4) to accurately model the commuters’ decision making (e.g., route choice).

We model the Singapore rail transit system in our experiments. The Singapore RTS comprises two train systems: the Mass Rapid Transit (MRT) system consisting of four

Computational model

Our approach to modelling the RTS comprises two aspects: (1) the modelling of the trains as they traverse the rail network, and (2) the modelling of commuters as they travel from their origins to their destinations. The first aspect, the modelling of trains, is straightforward as we do not fully model the physical mechanisms of the trains, and it is only sufficient that our simulated train arrivals can fit the publicly available train schedules (i.e., first-train timings and train arrival

Evaluation of agent-based model and simulation

Having validated the model in the previous section, here, we evaluate the model and determine if it can adequately replicate the mechanics of the rail-transit system. First, we investigate the effective train capacity, which can be lower than the operational capacity of the train. Second, we observe dynamics recorded in our simulation that are not explicitly manifested in the empirical data, including the penultimate station effect. Lastly, we explore how the penultimate station effect can be

Scenario descriptions

A major concern to urban planners is the scalability of their transit systems with regard to commuter population growth. In our population scaling scenarios, we adjust the transit demand in our model to predict how population growth may affect commuter experiences with respect to travel duration and number of trains missed. Here, population refers to the number of journeys simulated. An actual commuter may take multiple journeys in a single day. In these scenarios, we utilize the Monday

Conclusion

In this work, we had incorporated empirically derived travel demand data into a full-scale agent-based model of the train rapid transit system in Singapore. Our approach granted us a more comprehensive view of congestion dynamics than afforded by analysing the dataset directly. We were able to synthesize highly detailed measurements, including crowdedness and number of trains missed. With these measurements, transport operators can accurately estimate the comfort and satisfaction of commuters,

Acknowledgements

We thank the Land Transport Authority of Singapore for providing the anonymized smart card data. We thank Dr. Gary Lee Kee Khoon and Dr. Terence Hung Gih Guang for discussions. This research is supported by the Science and Engineering Research Council of the Agency for Science, Technology and Research (A*STAR) of Singapore (Complex Systems Programme grant number 122 45 04056); and by the A*STAR Computational Resource Centre through the use of its high performance computing facilities.

Nasri Bin Othman is a Research Engineer at the Institute of High Performance Computing, Singapore, where he investigates urban systems. He received his B.Eng. in Computer Science from Nanyang Technological University, Singapore. His research interests include modelling and simulation, urban transportation, evolutionary computation, and interactive visualization.

References (21)

M. Bagchi et al.
The potential of public transport smart card data
Transp. Policy
(2005)
J.K. Chung et al.
Measures of distance between probability distributions
J. Math. Anal. Appl.
(1989)
E.F. Legara et al.
Critical capacity, travel time delays and travel time distribution of rapid mass transit systems
Physica A
(2014)
K. Leyton-Brown et al.
Incentive mechanisms for smoothing out a focused demand for network resources
Comput. Commun.
(2003)
C. Morency et al.
Measuring transit use variability with smart-card data
Transp. Policy
(2007)
N.B. Othman et al.
Simulating congestion dynamics of train rapid transit using smart card data
Procedia Comput. Sci.
(2014)
M.P. Pelletier et al.
Smart card data use in public transit: a literature review
Transp. Res. Part C: Emerg. Technol.
(2011)
A. Bhattacharyya
On a measure of divergence between two statistical populations defined by their probability distributions
Bull. Calcutta Math. Soc.
(1943)
E. Bonabeau
Agent-based modeling: methods and techniques for simulating human systems
Proc. Natl. Acad. Sci. U.S.A.
(2002)
A. Chakirov et al.
Use of public transport smart card fare payment data for travel behaviour analysis in Singapore

There are more references available in the full text version of this article.

Cited by (16)

SUMMIT: A multi-modal agent-based co-simulation of urban public transport with applications in contingency planning
2023, Simulation Modelling Practice and Theory
In this article, we present SUMMIT (Singapore Urban Multi-Modal Integrated Transport Simulator), a multi-modal agent-based simulation platform for public transport calibrated using real world mobility data sets from Singapore. SUMMIT uses a co-simulation approach in order to model multiple modes of public transport in tandem with commuters seamlessly transitioning between different mode simulations. This is implemented through a message passing framework codenamed Fabric (Fast, Agent-Based, Reproducible, Integrated Co-simulation) which helps to align and synchronise the simulation as it transitions through the different simulation time steps. Three stand-alone simulators, each modelling a key mode in the Singapore public transport system, namely train, bus and taxi, are integrated with a commuter control model, which simulates the commuter behaviour and takes care of commuters’ transfer between different modes. We present one application of SUMMIT in public transport contingency planning, where train services are disrupted in multiple lines of Singapore Mass Rapid Transit (MRT) network. Different scenarios are simulated and key performance indicators are compared to help planners evaluate the efficacy of mitigation measures. The results show that bridging bus services, a commonly used mitigation measure, are effective in reducing crowd sizes within the train stations. However, we demonstrate that bridging buses could also result in worsening of commuters’ travel time due to over-demand for these buses. Our results also show early dissemination of information to commuters during a train service disruption could reduce the negative impacts of the disruption event significantly.
Real-time agent-based crowd simulation with the Reversible Jump Unscented Kalman Filter
2021, Simulation Modelling Practice and Theory
Citation Excerpt :
Recent efforts have been made to develop methods that will allow ABMs to react to real-world events as they happen. Whilst promising, they exhibit a number of limitations such as: the need for manual calibration [13] (which is infeasible in most cases); implementations that contain only a few agents and/or limited interactions [14,15]; the use of agent-based models that are simple enough to be approximated by an aggregate mathematical model [3,16]; parameters that can be dynamically optimised but a model state that is not updated [17]; or the use of DA methods that do not scale to large model size [10,15]. Also, importantly, none of the previous approaches have attempted to assimilate data for categorical variables.
Commonly-used data assimilation methods are being adapted for use with agent-based models with the aim of allowing optimisation in response to new data in real-time. However, existing methods face difficulties working with categorical parameters, which are common in agent-based models. This paper presents a new method, the RJUKF, that combines the Unscented Kalman Filter (UKF) data assimilation algorithm with elements of the Reversible Jump (RJ) Markov chain Monte Carlo method. The proposed method is able to conduct data assimilation on both continuous and categorical parameters simultaneously. Compared to similar techniques for mixed state estimation, the RJUKF has the advantage of being efficient enough for online (i.e. real-time) application. The new method is demonstrated on the simulation of a crowd of people traversing a train station and is able to estimate both their current position (a continuous, Gaussian variable) and their chosen destination (a categorical parameter). This method makes a valuable contribution towards the use of agent-based models as tools for the management of crowds in busy places such as public transport hubs, shopping centres, or high streets.
Challenges, tasks, and opportunities in modeling agent-based complex systems
2021, Ecological Modelling
Citation Excerpt :
As a result, it is almost impossible to reproduce model results or to interpret them. For instance, Othman et al. created an ABM of the rail network in Singapore, which used only train ticket purchasing data as its input (Othman et al., 2015). There is an implicit assumption that models built using new forms of micro-data will capture the essential processes that are taking place in these systems.
Humanity is facing many grand challenges at unprecedented rates, nearly everywhere, and at all levels. Yet virtually all these challenges can be traced back to the decision and behavior of autonomous agents that constitute the complex systems under such challenges. Agent-based modeling has been developed and employed to address such challenges for a few decades with great achievements and caveats. This article reviews the advances of ABM in social, ecological, and socio-ecological systems, compare ABM with other traditional, equation-based models, provide guidelines for ABM novice, modelers, and reviewers, and point out the challenges and impending tasks that need to be addressed for the ABM community. We further point out great opportunities arising from new forms of data, data science and artificial intelligence, showing that agent behavioral rules can be derived through data mining and machine learning. Towards the end, we call for a new science of Agent-based Complex Systems (ACS) that can pave an effective way to tackle the grand challenges.
Improved estimation of commuter waiting times using headway and commuter boarding information
2018, Physica A: Statistical Mechanics and its Applications
Citation Excerpt :
Many modern cities around the world have implemented Automated Passenger (fare) Collection (APC) data and Automated Vehicle Location (AVL) procedures that seek to improve the overall efficiency of bus services. In addition, the analysis of datasets resulting from these platforms have increased our understanding of the dynamics of bus operations [6–10]. Unfortunately, the dynamics of actual commuter waiting times remains difficult to accurately record.
The average amount of waiting time spent by commuters is one of the key indicators of service quality for public bus operations. While actual measurements of actual waiting time is difficult to be done en masse, models of waiting time can be derived from bus headways and these models have been adopted by transport planners in monitoring and regulating service reliability of operators. However, these models are founded on several assumptions on the patterns of commuter arrival which may not be applicable for bus services that experience high demand and heavily fluctuating commuter patterns. Given the availability of granular data on commuter boarding from automated fare collection systems, we propose a new methodology to better estimate the average waiting time of commuters. The formulation is anchored and validated using a three-month dataset from ten selected bus routes in Singapore. Finally, we discuss how our new measure allows for minimization of commuter waiting time through schedule optimization.
Modeling and evaluation of highly complex computer systems architectures
2017, Journal of Computational Science
Citation Excerpt :
In the design phase, modeling is generally adopted as a means to explore the solutions before developing prototypes: for HCCS, prototypes may be very expensive or not a viable possibility. For example, systems that must strictly adhere to safety standards and require a proof of conformance in order to be viable [4], such as railways or aerospace systems [5–7], in which models may help in reducing the use of expensive prototypes and lower the costs of the overall design phase; or, in the case of very large computing facilities, adopting a prototype in the small will not reveal all problems that may arise in the large scale, such as communication problems, energy consumption effects [8,9], workload interactions [10,11], scheduling effects at various scales [12–14], and all the adaptations to the evolutions of the workload in the lifespan of the system [15], and in this case models may help in choosing between alternative approaches that show their effects in the large only, after big investments, or may signal in advance emerging problems such as saturation, competition for resources or side effects. In the integration phase, the problem lies in the possible mismatches or, again, interactions between components, that may lead to unexpected problems.
Modern computer based systems are characterized by several complexity dimensions: a non-exhaustive list includes scale, architecture, distribution, variability, flexibility, dynamics, workloads, time constraints, dependability, availability, security, performances. The design, implementation, operation, maintenance and evolution of such systems require informed decisions, that must be founded onto techniques and tools enabling an anticipated knowledge about the behavior of every subsystem, including hardware, software and interactions, and the whole system, and the relationship of the system and the external world, considering workloads, communication, sensing of physical interactions. Performance prediction, and in general behavior prediction, may exploit simulation based approaches or analytical techniques to evaluate in advance the effects of design choices, or variability under different workloads, or emerging behavior, of systems, and provide a valuable support in all the phases of the lifecycle of a system by means of proper modeling approaches. In this Special Issue we present some contributions that offer a glance on modeling and evaluation of complex computer based system, and that have been chosen in order to provide a view on different domains and different approaches, mainly focusing on simulation techniques and related applications.
Perspectives of the International Conference of Computational Science 2014
2015, Journal of Computational Science
Computational Science has enabled a raft of science that was either impossible, dangerous, or extremely expensive. It is arguably one of the most multi-disciplinary research endeavors, and draws on foundational work in mathematics and computer science. Computational Science has applicability in almost all scientific domains, and is now an essential tool in many of these. This special section contains extended papers originally published in proceedings of the 14th International Conference on Computational Science (ICCS 2014), an annual event that promotes leading edge research.

View all citing articles on Scopus

Erika Fille Legara, a former scholar at the Santa Fe Institute Complex Systems Summer School, obtained a Ph.D. in Physics at the National Institute of Physics at the University of the Philippines, Diliman. Bulk of her work has been on the diverse applications of statistical mechanics and network theory in understanding the mechanisms behind various socio-economic systems and paradigms such as multi-level markets, telecommunication companies, and news and media framing. She is currently a Scientist at the A*STAR Institute of High Performance Computing (IHPC) working on big data and visualization, urban complexity, and complex networks. More details on her research can be found at: http://www.erikalegara.com

Vicknesh Selvam is a scholar with the Agency of Science, Technology and Research (A*STAR) currently working as a Research Engineer at the Institute of High Performance Computing (IHPC). He received his Bachelor of Sciences in Mathematics, with a specialization in Computing, at the University of California, Los Angeles. His research interests include Network Science, Image Processing, Cryptography and Mathematical Modelling. He is currently working on the data analysis and modelling of Urban Transport networks under the Urban Systems Initiative.

Christopher Pineda Monterola is a Senior Scientist at the Institute of High Performance Computing in A*STAR Singapore. He is currently the Capability Group Manager (CGM) of the Complex Systems (CxSy) Group at the IHPC. Chris is also the Principal Investigator of the Complexity Science Programme of the IHPC under the CxSy. Prior to his stint in Singapore, he was a postdoctoral fellow at the Max Planck Institute for the Physics of Complex Systems. He obtained a Ph.D. in Physics in 2002 from the National Institute of Physics, University of the Philippines Diliman (UPD) where he was an Associate Professor 7. More details on his works can be found at: http://www.chrismonterola.net

^☆: This article is an extension of [16].

View full text

A data-driven agent-based model of congestion and scaling dynamics of rapid transit systems☆

Highlights

Abstract

Introduction

Section snippets

Data

Computational model

Evaluation of agent-based model and simulation

Scenario descriptions

Conclusion

Acknowledgements

Transp. Policy

J. Math. Anal. Appl.

Physica A

Comput. Commun.

Transp. Policy

Procedia Comput. Sci.

Transp. Res. Part C: Emerg. Technol.

On a measure of divergence between two statistical populations defined by their probability distributions

Bull. Calcutta Math. Soc.

Agent-based modeling: methods and techniques for simulating human systems

Proc. Natl. Acad. Sci. U.S.A.

Use of public transport smart card fare payment data for travel behaviour analysis in Singapore