An exploration of data-driven microscopic simulation for traffic system and case study of freeway

Traffic simulation systems have been widely used for traffic system analysis and optimization. The traditional simulation system uses analytical models for each kind of driving behaviour, with little regard for the coupling relationship between the models, resulting in limited performance and complicated calibration and validation processes. In this study, a Data-Driven Simulation System (DDSS) framework was introduced to define traffic system operation processes and coordinate submodules. The proposed DDSS includes a General Kernel for running the simulation environment, and a Customized Interface for accessing to various data-driven driving behaviour models. To unify the modelling of multiple driving behaviour and increase prediction accuracy, a data-driven Sim-Hybrid Retraining Constrained LSTM (SHRC-LSTM) model was built. In addition, experiments on two NGSIM freeway testbeds demonstrated that DDSS produced more precise results in terms of efficiency, safety, and emissions than the most widely-used simulation system VISSIM.

Automated summary

Traditional traffic simulation systems have limitations in performance and calibration/validation processes due to their use of independent models without considering their coupling relationship. In this study, a Data-Driven Simulation System (DDSS) framework was introduced to address this issue by defining traffic system operation processes and coordinating submodules. Experimental results showed that DDSS outperforms the widely used VISSIM simulation system in terms of accuracy and overall performance.