A hybrid deep learning approach for urban expressway travel time prediction considering spatial-temporal features

Travel time is an effective measure of roadway traffic conditions which enables travelers to make smart decisions about departure time, route choice and congestion avoidance. Recent years have witnessed numerous successes of deep learning neural networks in the domains of artificial intelligence (AI). Motivated by the dominant performance of convolution neural networks (CNNs) and long short-term memory neural networks (LSTMs), and with consideration of the spatial-temporal features, this study attempts to develop a hybrid deep learning framework fusing CNNs and LSTMs to forecast the travel time on urban expressways. A 2-dimension deep CNNs is exploited to capture spatial features of traffic states, and LSTMs are utilized to excavate the temporal correlation of travel time series. Then, these spatial-temporal features are fed into a linear regression layer. The travel time forecasting is achieved by fusing these abstract traffic features in a hybrid deep learning framework. The proposed approach is investigated on Ring 2, a 33km urban expressway of Beijing, China. The results demonstrate the advantage of the proposed method, as well as its feasibility and effectiveness compared with other prevailing parametric and nonparametric algorithms.