The Development and Calibration of an Agent-Based Microsimulation Model for Vehicle-Pedestrian Interaction

With the increases in computer processing power and advances in programming skills, an array of transportation and urban planning computer models are now available to the profession. They are extensively used in developed nations to model complex transport scenarios and interactions. Models vary from a macroscopic level, which focus on the system as a whole and a higher aggregation level, to more complex microscopic tools, which allow the simulation of individual road users and behaviour to obtain more realistic representations at a local/street level. Microsimulation models, to varying degrees, give researchers and practitioners the ability to analyse the effectiveness of interventions on a disaggregated level, as individual vehicles and/or pedestrians are simulated in detail as they move through the road network with the goal of reaching their destination by the most cost effective or shortest route. The interaction of road users in microsimulation models is refined via parameter settings of, for instance, compliance levels (i.e. attitude to risk), walking speeds, aggression and awareness levels, etc. Default values of these parameters are incorporated in the software at levels set by the vendors. These values affect outcomes of modelled scenarios and may not replicate actual road user behaviour. Calibration of the model is, therefore, vital to match observed conditions and to produce realistic outputs. However, the complexity of microsimulation models and the large number of parameters that require specification means that calibration is a complex and iterative process. Furthermore, direct measurement of parameter values is very difficult because many of them represent subtle features that are hard to isolate and, because of the extensive amount of data collection required at a disaggregate level. This paper describes the development of a simulation model and the evaluation of parameters for a study of road user interaction for a local arterial road in Cape Town, South Africa, using a commercially available package, Paramics. Results are compared to values obtained using default settings, which are drawn from developed world contexts. Visual reviews of the calibrated network reveal that, although the software allows flexible modelling of shared vehicles and pedestrian space, it does not replicate observed local road user behaviour of some vehicles and pedestrians at crossings. Given this finding, an investigation into the possibility of modifying appropriate parameters is also presented.