Pieter J. Fourie

Large-scale agent-based combined traffic simulation of private cars and commercial vehicles

The number of independent and interdependent freight actors (firms), the complex supply chain structures between them, and the sensitivity of shipment data are some reasons that the modeling of freight traffic is lagging its public and private transit counterparts. An agent-based approach was used to reconstruct commercial activity chains and simulate them, along with private vehicles, for a large-scale scenario in Gauteng, South Africa. The simulated activities were compared with the actual observed activities of 5,196 vehicles that were inferred from Global Positioning System logs covering approximately 6 months. The results show that the reconstructed activity chains are both spatially and temporally accurate, especially in areas of high activity density. Freight vehicles are a major contributor to traffic congestion and emissions, and the new approach bridges the gap between the person and the state of the art in commercial transport modeling.

Transport modelling in the age of big data

ABSTRACT New Big Data sources such as mobile phone call data records, smart card data and geo-coded social media records allow to observe and understand mobility behaviour on an unprecedented level of detail. Despite the availability of such new Big Data sources, transport demand models used in planning practice still, almost exclusively, are based on conventional data such as travel diary surveys and population census. This literature review brings together recent advances in harnessing Big Data sources to understand travel behaviour and inform travel demand models that allow transport planners to compute what-if scenarios. From trip identification to activity inference, we review and analyse the existing data-mining methods that enable these opportunistically collected mobility traces inform transport demand models. We identify that future research should tap on the potential of probabilistic models and machine learning techniques as commonly used in data science. Those data-mining approaches are designed to handle the uncertainty of sparse and noisy data as it is the case for mobility traces derived from mobile phone data. In addition, they are suitable to integrate different related data sets in a data fusion scheme so as to enrich Big Data with information from travel diaries. In any case, we also acknowledge that sophisticated modelling knowledge has developed in the domain of transport planning and therefore we strongly advise that still, domain expert knowledge should build the fundament when applying data-driven approaches in transport planning. These new challenges call for a multidisciplinary collaboration between transport modellers and data scientists.

Synthetic Population Generation by Combining a Hierarchical, Simulation-Based Approach with Reweighting by Generalized Raking

A recent approach for generating populations of synthetic individuals through simulation is extended to produce households of grouped individuals. The contingency tables of the generated populations match external controls on the individual and household levels while exhibiting far greater variety in composition than existing approaches can offer. The method involves a two-step approach. The first consists of a procedure based on Gibbs sampling, which has only recently been applied to population generation in transportation modeling and is generically called Markov chain Monte Carlo (MCMC). For this work, the model was generalized, and an extension was developed, hierarchical MCMC, which was able to generate a hierarchical structure. The second step, a postprocessing step, uses generalized raking (GR), which reweights the output from hierarchical MCMC to perfectly satisfy known marginal control totals on the individual and household levels. The application input data—a demographic sample and some known marginals from Singapore—added further complexities to the problem, which had not yet been explored in the current literature. Despite data challenges, consecutively applying the methods above produced realistic synthetic populations. Results confirm their goodness of fit and their generated hierarchical structures.

Increased Convergence Rates in Multiagent Transport Simulations with Pseudosimulation

A multimodeling approach to large-scale, activity-based, multiagent simulation of travel demand is introduced. MATSIM is a full activity-based transport simulation. Its greatest current performance limitation is the network loading simulation, currently a queue simulation (QSim). QSim is iteratively executed for the entire agent population for evaluating the effects of random mutations on the activity plans of a fraction of the population. After each QSim, poorly performing plans are discarded, good plans are kept, and the agents slowly learn what works best for their individual activity needs. In the application presented, the system periodically replaces QSim for a number of iterations with a simplified pseudosimulation that runs approximately two orders of magnitude faster. The pseudosimulation uses travel time information from the preceding QSim iteration to estimate how well an agent day plan might perform. Repeated iterations of the pseudosimulation produce better-performing plans in a short time. These plans are passed to the QSim for updating of network travel time information, and the process repeats. The technique is tested in a scenario for Zurich, Switzerland, and incorporates mode choice, road pricing, secondary activity location choice, activity timing adjustment, and dynamic routing. The technique dramatically improves convergence rates for such complex, large-scale simulations and fully exploits modern multicore computer architectures.

Simulation of autonomous transit on demand for fleet size and deployment strategy optimization

Abstract Autonomous transit on demand (ATOD) is a potential future public transit mode, which appeals to a lot of researchers and policymakers. In the project, ATOD is simulated in MATSim to explore the optimal fleet size and deployment strategy to help policymakers to decide how to introduce the new transport system in the future. The simulation enables the system to explore the optimization automatically under specific constraints with the MATSim evolutionary algorithm.

Multi-agent urban transport simulations using OD matrices from mobile phone data

Abstract Although new available big data sources have revealed themselves to be extraordinarily useful for transport demand modelling, they have not come into widespread use due to the justifiable privacy concerns of data stewards. In this study, we step back and re-evaluate the way in which mobile phone telco data can be introduced for the task of transport and land-use policy evaluation, travel demand forecasting and transport infrastructure testing through large-scale transportation simulations. We investigated that question by deploying a multi-agent transport simulation driven primarily by hourly-aggregated telco Origin-Destination (OD) matrices. We address the principal four challenges: spatial and temporal disaggregation, mode imputation and route choice. For temporal disaggregation, we propose a convolution with an exponential kernel method. As for transport mode imputation, a supervised-learning framework is designed. The simulation results are compared against traffic count data and public transport smart card transactions, showing accurate patterns for private cars but overestimated public transport demand in the morning peak. Lastly, we set the future steps for the improvement of simulations driven by aggregated mobile phone data.

Interactive Analysis and Decision Support with MATSim

Agent-Based Simulation Means Lots of Data Agent-based transport demand models require managing and integrating data sources several orders of magnitude larger than traditional aggregate models. In a truly disaggregate demand description, as seen in our MATSim implementation for Singapore, spatial data represents individual buildings and land parcels, not zones; travel demand takes the form of a full activity diary with connecting trips for every individual, based on their personal demographic attributes, instead of an aggregate number of trips from zone to zone for a speci c time period. For this reason, input data for an aggregate four-step (or related) demand model can generally be edited on a laptop, using standard spreadsheet so ware, whereas agentbased modeling requires the manipulation and synthesis of large stores of structured, hierarchical data, frequently exceeding most personal computer capacity.