Route Planning by Merging Local Edges into Domains with LSTM
Xiaolong Chen, Huipeng Zhang, Fei Xiao, Dajing Peng, Chuanming Zhang, Bo Hong · 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC) · 2022
With the development of positioning technology, massive trajectories are recorded. How to effectively use hidden features in massive trajectories for route planning has attracted widespread attention. Based on the distribution characteristics of the trajectories and the modeling ability of the RNN for variable long-term sequences, we summarize the trajectory distribution through the deep learning model. However, due to the excessive length of the original trajectory sequence, the direct use of the original trajectory to train the RNN causes large search space. In this paper, the sequences in the original trajectory are merged according to the road name, which effectively reduces the search space and speeds up inference. At the same time, the source and destination direction, grid information, attributes of road, and other information are added to improve the inference effect. Finally, the possible routes between the source and destination (SD pair) are output by beam search, which can be ranked according to the distance and travel time to achieve multi-route recommendations. Our experiment based on real travel trajectories shows that our approach outperforms the existing approaches.