关键词: 长短期记忆模型；神经网络；跟车距离

Abstract: Many forward collision warning systems currently rely on warning distance as a feature to alert drivers, so improving the accuracy of predicting following distance can effectively enhance the warning capability of such systems. In this study, a driving simulator is used to construct following scenarios, and behavior data of 41 drivers are collected. The experimental data are divided into training and testing sets in a 3:1 ratio. By using the drivers' following distance and speed as inputs to a long short-term memory model, the following distance is predicted and analyzed. The results show that the model using this dataset could accurately predict the drivers' following behavior, with good generalization performance and no overfitting. Furthermore, by testing different time window lengths, it is found that as the length of the testing set decreased, the prediction error increased. This study provides theoretical support for improving the accuracy of forward collision warning systems, thereby increasing drivers' acceptance of these systems.

Key words: Long short-term memory model; Neural network; Following distance