Revisit of Microscopic Car Following Models: Conventional and Machine Learning Perspectives

Publication Type:
Issue Date:
Full metadata record
Microscopic car following models, or simply car following models, are used to determine how vehicles are following one another on roadways. They are the foundation of microscopic traffic flow theories and are of great importance with regard to the developments of adaptive cruise control (ACC) system and connected and automated vehicles (CAV), as well as the evaluation of intelligent transportation system (ITS) strategies. Therefore, it is very demanding to keep improving existing car following models or develop new ones towards better reproducing human driver behaviours or addressing specific traffic challenges. In this thesis, we mainly focus on conventional, mathematical-equation based car following models and novel, machine-learning based car following models. The aims and purposes of this thesis consist of five folds: 1) to take intensive revisits into several widely-used microscopic car following models (from conventional model family or machine-learning based model family) to further study their advantages and deficiencies; 2) to propose corresponding solutions (by modifying car following model structure or model calibration approach) towards addressing such deficiencies of existing models; 3) to develop new car following models using novel machine learning technologies to solve traffic challenges that human drivers are hard to overcome; 4) to comprehensively evaluate the performances of the proposed car following models by comparing them with similar, existing counterparts from various aspects; 5) to discuss about the application fields where these proposed car following models can be best applied.
Please use this identifier to cite or link to this item: