Data-driven Modeling of Car-Following Behavior on Freeways Considering Spatio-Time Effects: A Comparison of Different Neural Network Structures

Car-following behavior models based on conventional mathematical cannot adequately reproduce traffic phenomena, such as breakdown, capacity drop, and oscillations, require parameter setting. Therefore, this study aims to construct a highly accurate car-following model using deep learning. We evaluated the influence of variables in dataset random forest. Furthermore, we constructed predict acceleration one second learning methods, neural network, long short-term memory, one-dimensional convolution network (1DCNN), 2DCNN models. The were root mean square error, MAE, yyplot, loss plot. results showed that spatiotemporally structuring data increased accuracy predictions.