Open AccessA Method for Traffic Flow Forecasting in a Large-Scale Road Network Using Multifeatures
Author(s) -
Chuhao Zhou,
Peiqun Lin,
Xukun Lin,
Yang Cheng
Publication year - 2021
Publication title -
promet
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.315
H-Index - 19
eISSN - 1848-4069
pISSN - 0353-5320
DOI - 10.7307/ptt.v33i4.3709
Subject(s) - computer science , robustness (evolution) , artificial neural network , backpropagation , benchmark (surveying) , data mining , scale (ratio) , traffic flow (computer networking) , artificial intelligence , machine learning , geography , biochemistry , chemistry , cartography , computer security , geodesy , gene
Accurate traffic prediction on a large-scale road network is significant for traffic operations and management. In this study, we propose an equation for achieving a comprehensive and accurate prediction that effectively combines traffic data and non-traffic data. Based on that, we developed a novel prediction model, called the adaptive deep neural network (ADNN). In the ADNN, we use two long short-term memory (LSTM) networks to extract spatial-temporal characteristics and temporal characteristics, respectively. A backpropagation neural network (BPNN) is also employed to represent situations from contextual factors such as station index, forecast horizon, and weather. The experimental results show that the prediction of ADNN for different stations and different forecast horizons has high accuracy; even for one hour ahead, its performance is also satisfactory. The comparison of ADNN and several benchmark prediction models also indicates the robustness of the ADNN.