CNN-LSTM Based Traffic Prediction Using Spatial-temporal Features
Zhen Zhao, Ze Li, Fuxin Li, Yang Liu
Abstract
Aiming at the problem of traffic congestion prediction based on taxi big data, a CNN-LSTM based traffic prediction model using spatial-temporal trajectory topology is proposed. First, the trajectory information is abstracted into a trajectory topology map with spatial-temporal characteristics according to the time and space dimensions. The topology map solves the problem that the road network map does not have stationarity, and extracts a variety of road condition influence factors. Then, the spatial characteristics of the trajectory traffic flow are extracted by CNN, and the temporal characteristics of the trajectory traffic flow are extracted according to the memory characteristics of LSTM. The experimental results show that the RMSE, MAPE and Spearman correlation coefficients of the proposed method on JT-T809-2011 dataset have an absolute improvement of 1%~2% over state-of-the-art strategies.