Forecasting transportation flow is of vital significance for relieving traffic congestion and improving public safety. However, it is very challenging to achieve it precisely because many factors such as weather condition, traffic control and big celebration events can lay great influence on it. To better fulfill this challenging task, we propose a deep-learning-based approach called Spatio-Temporal Convolutional Neural Network. We first model three temporal properties of transportation flow (closeness, period, trend). Each property is assigned with a convolutional neural network, each of which models the corresponding property of public traffic. This model also fuses the aggregation of the output of the three properties with external elements, for example weather condition and some big events, to gain a better performance in citywide traffic flow prediction. Experiments on Beijing taxi flow and the New York city bike flow show that our ST-CNN model outperforms many well-known passenger flow prediction methods.
Deep Spatio-Temporal Convolutional Neural Network for City Traffic Flow Prediction
2021-01-01
584035 byte
Conference paper
Electronic Resource
English
Deep neural network robust traffic prediction method based on multi-modal spatio-temporal data
European Patent Office | 2021
|Spatio-Temporal AutoEncoder for Traffic Flow Prediction
IEEE | 2023
|European Patent Office | 2020
|