A cellular automata traffic flow model combined with a BP neural network based microscopic lane changing decision model

The purpose of this study is to propose a cellular automata (CA) traffic flow model with high accuracy for lane change decision and name it LCCAM. A driving simulator experiment was conducted to find factors affecting lane changing decisions. A back-propagate (BP) neural network was used to obtain the lane changing rules for the microscopic lane changing decision model (LCDM), and the collected accurate vehicular trajectory data were used to train the BP neural networks for the prediction of lane changes. After comparing different input variable combinations, the most accurate input setting was determined, including the locations and velocities of neighboring vehicles, inner/outer lane indicator, and the speed limits of the corresponding lane. Then, the determined BP neural network was adopted in the LCCAM as the LCDM. Simulation results showed that the LCCAM can capture important characteristics such as the mean velocities and the number of lane changes well, by comparing with observed traffic flow. Meanwhile, the LCCAM illustrates a better performance in replicating the number of lane changes than the other reference CA models. The research results show that the LCCAM proposed in this study will have potential and value for autonomous driving and active safety analysis in the future.