A Bidirectional Context-Aware and Multi-Scale Fusion Hybrid Network for Short-Term Traffic Flow Prediction

Short-term traffic flow prediction is to automatically predict the changes in a period of future time based on extraction spatiotemporal features road network. For governments, timely and accurate crucial plan manage-ment improve efficiency. Recent advances deep learning have shown their dominance short-term prediction. However, previous methods are mainly limited temporal so far failed bidirectional con-textual relationship correctly. Besides, precision practicality by network scale single scale. To remedy these issues, Bidirectional Context-aware Multi-scale fusion hybrid Network (BCM-Net) proposed, which novel framework changes. In BCM-Net, (BCM) block added feature structure effective-ly integrate features. The Interpolation Back Propagation sub-network used merge multi-scale information, further improves robustness model. Experiment results diverse datasets demonstrated that proposed method outperformed state-of-the-art methods.