STRMA：Spatial Temporal Reversal Memory Autoencoder for Traffic Flow Prediction

Abstract

Urban trajectory data contains rich semantic and behavioral information, among which traffic trajectory data is a key technical support in intelligent transportation systems. Therefore, this article focus on solving traffic flow prediction.

Currently, most methods employ temporal and spatial modeling to capture traffic flow correlations by aggregating historical information across time and space dimensions. This paper proposes a different approach to traffic flow prediction. Specifically, we introduce a novel spatial-temporal encoder, STRMA (Spatial Temporal Reversal Memory Autoencoder), which captures correlations in both temporal and spatial dimensions of the input traffic flow data to predict future traffic flow.