Abstract: In order to further fully extract the spatiotemporal and periodic characteristics of traffic flow, this paper adopts a combined model (DLSTM-AE) of autoencoder architecture and Deep Long Short-Term Memory neural networks ...