Analysis of different approaches for future land transport technologies and emissions

Kurzfassung

With an increasing need to shift towards sustainable modes of transport, electric vehicles are already emerging as a viable alternative to conventional fuel vehicles. However, there is a wide regional diversity in the market penetration rates of such alternate fuel vehicles. Hence models which can reflect the complex and varied consumer demands and needs must be developed as a reference for relevant stakeholders to make informed decisions and devise suitable policy. This thesis explores data-driven models to forecast future market shares of alternate vehicles under possible scenarios. The spatial and temporal evolution of passenger electric vehicles was done using machine learning algorithms by generating a synthetic population and predicting its resulting market potential and sales on a yearly and zip-code level basis. Household types were clustered via K-means algorithm to identify different consumer segments. These were then distributed spatially based on income and region-type covariates using Gaussian Mixture model methods. By using mobility surveys for German households containing only 350 EV owning households from around 81000 households, the model was able to quantify sensitivity to ownership potential and recognize patterns in consumer behaviors. Age and household income were observed as important factors affecting a consumer’s willingness to pay more during the transient EV penetration period. For eg., younger populations were found to be willing to spend 32 % more than their affordable price range for an Electric vehicle. Market share for future scenarios that were defined based primarily on charging infrastructure availability and vehicle costs, were analyzed. The housing split distribution of consumer segments dictates the availability of daily charging infrastructure and the effect of this was observed in the rate of EV market share development. The model was tested on France which reflected higher market potential than Germany due to higher incentives in the initial period. Region level comparison with actual EV sales for France gave a promising outlook. Despite imbalanced datasets to learn from and assumptions related to charging infrastructure availability, the model performs relatively well and is expected to improve with the inclusion of additional features related to technological improvements and other spatial covariates.