TrafficPro:a framework to predict link speeds on signalized urban traffic network
When the traditional deep learning-based models predict link speeds for the entire urban traffic network,they do not consider the proactive feature(signal control information)of traffic flow,and therefore achieve low prediction accuracy.In order to tackle this issue,this paper proposed a link speed prediction framework,based on generative adversarial network and graph neural network.By adopting a proactive and a reactive prediction module,the generator of this framework is able to encode traffic flow and signal control information at the entire network level.The discriminator is then used to increase the generalizability of the prediction outcome.By comparing its performance with traditional time-series and deep learning-based models in real-world traffic circumstances,it is found that the proposed framework achieved less prediction error(3％-5％RMSE drop)than the SOTA model(ASTGCN).
万方数据
维普
