Urban railway transit is of great significance in the daily lives of Metropolitan residents. Therefore, forecasting rail- way traffic is fundamental to urban management. However, very few research has been focused on collectively forecast railway transit in a citywide scale. With the development of location based service, the huge volume of GPS trajectory data make it possible for a citywide prediction of railway traffic. In this paper, we propose a deep-learning-based system named DeepRailway to predict and simulate rail- way traffic through heterogeneous data sources. Our data sources include huge volume of trajectory data and rail- way network. In our system, we firstly match the trajectory points to the railway network. And then the patterns of these trajectories are found using a network-based kernel density estimation (KDE), which converts the forecasting task into a sequence prediction problem. An LSTM recurrent neu- ral network model is built to predict the densities through- out the whole network. We evaluate our system in differ- ent timespan and prediction steps to verify its performance against other prediction methods.
DeepRailway: A Deep Learning System for Forecasting Railway Traffic
2018-04-01
2425719 byte
Conference paper
Electronic Resource
English
Empirical forecasting of railway traffic
Engineering Index Backfile | 1940
|Forecasting future volume of railway traffic
Engineering Index Backfile | 1924
|Forecasting future volume of railway traffic
Engineering Index Backfile | 1924
|A Causal Deep Learning Framework for Traffic Forecasting
IEEE | 2023
|