Development of vehicle emissions models for Australian conditions

Abstract

Microscopic simulation models such as AIMSUN, VISSIM and/or PARAMICS have the ability to output emissions based on default values for emission factors derived mainly from European test data. Emission algorithms in those models are based on overseas vehicle emissions datasets, which do not reflect the different Australian vehicles, fuels, climate and fleet composition. The proposed research provides a set of emission algorithms to be used in conjunction with traffic simulation modelling, to better represent local conditions. Macro level models based on average vehicle speeds may not be appropriate for use at a more localized and detailed level when vehicle speed profiles may change significantly. Emission rates for a number of vehicles were compared using Australian data based on dynamometer testing. The results show that only CO2 shows a strong correlation with average speed. All other pollutants show very low levels of correlation. On the other hand, an evaluation of several micro level emissions models has been undertaken by applying them to the results of Australian vehicle emissions measured in the field. A number of models have been analysed and the results compared. Power based models have some significant shortcomings and their use is not consistent with our finding that there are significant variations in emissions for small changes in vehicle power. The results highlight the need to model acceleration, deceleration and cruising stages of the urban cycle separately. A speed based approach, such as that followed in by the AIMSUN traffic simulation model, was found to have merit based on the evaluation results discussed here. The current research investigates the gap between estimated emissions and actual measurements using an Australian emissions dataset and the widely used micro-simulation model AIMSUN. The results indicate that the model adequately predicts CO2 emissions. However, the likely errors associated with the prediction of other pollutants are significantly greater. As a result, the thesis puts forward improved emissions estimation relationships for use with micro-simulation models. Using Australian emissions data, it was possible to improve the estimation ability of existing micro-simulation models. The thesis discusses the limitations of existing emissions estimation approaches at the micro level. A methodology to establish emission models for predicting emission pollutants other than CO2 is proposed. The models adopt a genetic algorithm approach to select the predicting variables. The approach is capable of solving combinatorial optimisation problems. Overall, the emission prediction results reveal that the proposed new models outperform conventional equations. There is a need to match emission modelling estimation to the accuracy levels of confidence in the outputs of transport models. In order to quantify the likely level of uncertainty attached to forecasts of emissions, an analysis of errors needs to be undertaken. The two major sources of error are the deficiency inherent in the model structure itself and the uncertainty in the input data used. This thesis deals with both of these error types in relation to CO2 emissions modelling using a case-study from Brisbane, Australia. To estimate input data uncertainty, an analysis of different traffic conditions using Monte Carlo simulation is shown here. Model structure induced uncertainties are also quantified by statistical analysis for a number of traffic scenarios. To arrive at an optimal overall CO2 prediction, the interaction between the two components was taken into account. Since a more complex model does not necessarily yield higher overall accuracy, a balanced solution needs to be found. The results obtained suggest that the CO2 model used in the analysis produces low overall uncertainty under free flow traffic conditions. However, when average traffic speeds approach congested conditions, there are significant errors associated with emissions estimates. Using different scenarios for different road configurations and traffic conditions, the results of applying the new approach are compared with those obtained by using default emissions parameters commonly found in a simulation package. The enhancement of emission predictions rests to a large extent on the further improvements to traffic micro-simulation models. The results obtained suggest that the new approach produces low overall errors under several traffic conditions. The accuracy of emissions predictions is, to a large extent, dependent on the errors associated with transport model outputs and on the accuracy of the emissions models themselves.