Assessing the uncertainty in micro-simulation model outputs

Micro simulation traffic models continue to be widely used as tools to analyse current and future road system performance. Such models tend to operate by concentrating on the movement of individual vehicles in the traffic stream on a second by second basis. The likely errors associated the outputs of such models are usually thought to be significantly lower than the 'coarser' models with deal with the movement of groups of vehicles over an average time period. This paper puts forward a methodology to evaluate the uncertainty in the outputs of micro level models by comparing those outputs with field data obtained from loop detectors. The paper concludes that model uncertainty shows very different characteristics for different traffic flow conditions. The Taguchi statistical method was applied to three traffic flow stages to establish the relationships between modelled vehicle speeds and measured speeds, on a minute-by-minute basis. The results show the close agreement between modelled and actual speeds under both congested and free flow conditions. In contrast, the transitional stage between free flow and queuing shows a higher level of discrepancy