An Urban Traffic Forecasting Model Based on Convolutional Neural Networks

• 1、Beijing Key Laboratory of Traffic Data Analysis and Mining, School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044

Abstract：In recent years, intelligent traffic control has becoming an important and challenging research topic. One of its key issues is the traffic forecasting problem.In this paper, we propose a model based on convolution neural networks to forecast citywide traffic state, namely, spatiotemporal convolutional networks (STCN). First, we designe a proper spatiotemporal data structure to store historical traffic state. Then, some necessary pre-processing is performed on the data. By analyzing spatiotemporal properties of traffic state data, we employ convolution to capture the spatial dependencies of traffic state and use three components to model the temporal dynamics.An experiment on a realworld Beijing traffic state data set demonstrates that STCN outperforms other 4 baseline methods.

Keywords： Convolution neural network Temporal and spatial property Urban traffic forecasting