Development and Evaluation of A Framework for Linking Traffic Simulation and Emission Estimation Models

Abstract

The need to understand the effect of policy decisions on environmental indicators is strong. The emergence of new technologies brought about by connected vehicle technologies, which are difficult to evaluate in field settings, means that policies must often be evaluated with software models. In these cases, however, the transportation model and the emissions model are often separate, and multiple different ways to connect these models are possible. Although the estimations provided by each model will vary, each method also differs in terms of the computational time.

This research is motivated by the need to understand the consequences of choosing a particular method to link a traffic and emissions model. Within the literature, aggregated approaches that simply use average speeds and volumes are often selected for their convenience and lower data needs. A number of different scenarios were therefore constructed to compare the estimates of these aggregated approaches to other methods that use disaggregated data, such as the use of individual discrete trajectories, the use of a velocity binning scheme that characterises networks based on their velocity profile or the use of a clustering algorithm developed for this study. This research presents a clustering algorithm that can be used to reduce the computational loads of an emissions estimation process without loss of accuracy.

The results of the analysis highlight the consequences of choosing each approach. Aggregated approaches produce unreliable estimates as they are backed by assumptions that may not be valid in every case. Using individual trajectories creates high computational loads and may not be feasible in all cases. The wealth of data available from a traffic microsimulation mean that using an aggregated approach neglects to utilise the full potential of the model; however, the hybrid approaches presented in this research (clustering and velocity binning) were found to make excellent use of this data while still minimizing computational demands.