In this paper, a data-driven car-following model based on the long short-term memory (LSTM) neural network in deep learning is proposed. Different from the traditional theoretical model, it can portray the complex decision-making and execution behaviors of drivers more accurately. The speed difference and the spacing between the leading vehicle and the following vehicle are set as the model input while the speed of the follower is set as the output. We determine the optimal values of parameters such as training times and the proportion of data used for testing and training and complete the modeling, testing, and optimization. Compared with other models, this model has two distinctive properties: First, it takes time-series data into consideration, which means the micro-behaviors of drivers such as the predictive ability and memory effects are integrated with the macroscopic traffic flow model. Second, we compared this model with models based on other neural networks such as SVR, FNN, and RNN. Respectively, the error value of this model is 83.8%, 69.5%, and 44.2% lower than those of the other three, which indicates that the car-following model based on LSTM has higher simulation accuracy than other kinds of the car following models.
Data-Driven Car-Following Modal Based on LSTM Deep Learning Method
01.03.2022
1440488 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Data-Driven Nonlinear Modal Analysis: A Deep Learning Approach
British Library Conference Proceedings | 2022
|British Library Conference Proceedings | 2022
|Deep reinforcement learning-based ethical-driven multi-modal decision-making method
Europäisches Patentamt | 2023
|A LSTM based Deep Learning Model for Text Summarization
IEEE | 2022
|