STTF: An Efficient Transformer Model for Traffic Congestion Prediction

Abstract With the rapid development of economy, sharp increase in number urban cars and backwardness road construction lead to serious traffic congestion roads. Many scholars have tried their best solve this problem by predicting congestion. Some traditional models such as linear nonlinear been proved a good prediction effect. However, with increasing complexity network, these can no longer meet higher demand without considering more complex comprehensive factors, spatio-temporal correlation information between In paper, we propose index devise new model transformer (STTF) based on transformer, deep learning model. The comprehensively considers speed segments, network structure, sections so on. We embed temporal spatial into through embedding layer for learning, use attention module mine hidden within data improve accuracy prediction. Experimental results real-world datasets demonstrate that proposed significantly outperforms state-of-the-art approaches.