Study on a Vehicle-Type-Based Car-Following Model using the Long Short-Term Memory Method

For car-following models, the car-following characteristics differ depending on the vehicle type, such as passenger cars, motorcycles, and trucks. Therefore, constructing a model for each category is essential. To that end, various modeling methods have been proposed; however, herein, we particularly focused on the long short-term memory (LSTM), which is the best method for forecasting long-term time-series data.[1, 2] The objective of this study was to construct a car-following model for each vehicle category using the LSTM and to evaluate the model accuracy for each vehicle category. In this study, US-101 and I-80 data provided by the next-generation simulation (NGSIM), which is based on natural traffic flow data, were used. In the NGSIM, only car-following situations were selected as car-following data, and these were classified into the vehicle type: motorcycles, passenger cars, and trucks. The classified data were then used to construct LSTM-based car-following models for each category. As the evaluation of this model was dependent on the amount of data, the model was built according to the amount of data for motorbikes, which was the lowest. A merged integrated model was also constructed by mixing the data obtained from all vehicle type, and the accuracy of the model was compared with that of the model for each vehicle category. Thus, this study investigated the impact of car-following models constructed for each vehicle category on the model accuracy using the LSTM. The results showed that for passenger cars and motorcycles, the RMSE values of the vehicle-specific models were smaller than those of the integrated model and the accuracy of the models was better than that of the integrated model. In future, we intend to construct car-following models for each vehicle characteristic.


  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01879972
  • Record Type: Publication
  • Source Agency: SAE International
  • Report/Paper Numbers: 2023-01-0680
  • Files: TRIS, SAE
  • Created Date: Apr 20 2023 9:56AM