PROJECT TITLE :

Fine-Grained Trajectory-based Travel Time Estimation for Multi-city Scenarios Based on Deep Meta-Learning

ABSTRACT:

Travel Time Estimation, also known as TTE, is an essential component of an intelligent transportation system (ITS). It is important to achieve the fine-grained Trajectory-based Travel Time Estimation (TTTE) for multi-city scenarios, which means to accurately estimate the travel time of the given trajectory for multiple city scenarios. This goal is significant. However, there are many obstacles to overcome, including dynamic temporal dependencies and fine-grained spatial dependencies, which contribute to the complexity of the factors. We propose a meta learning-based framework that we call MetaTTE. Its purpose is to continuously provide accurate travel time estimation over time by leveraging a well-designed deep neural network model that we call DED. DED is comprised of a Data preprocessing module and an Encoder-Decoder network module. The goal of this framework is to combat the challenges described above. The generalization ability of MetaTTE is enhanced using a small amount of examples thanks to the introduction of meta-learning techniques. This opens up new opportunities to increase the potential of achieving consistent performance on TTTE when traffic conditions and road networks change over time in the future. A network of encoders and decoders is utilized by the DED model in order to capture representations of finer-grained spatial and temporal details. Extensive experiments are performed on two real-world datasets in order to confirm that our MetaTTE outperforms nine state-of-the-art baselines. The results show that our MetaTTE improves accuracy by 29.35% on the Chengdu dataset and by 25.93% on the Porto dataset when compared to the best baseline.