Wide-Attention and Deep-Composite Model for Traffic Flow Prediction in Transportation Cyber–Physical Systems

摘要: Recently, traffic flow prediction has drawn significant attention because it is a prerequisite in intelligent transportation management urban informatics. The massively available data collected from various sensors cyber–physical systems brings the opportunities accurately forecasting trend. Recent advances deep learning shows effectiveness on though most of them only demonstrate superior performance single type vehicular carriers (e.g., cars) and does not perform well other types vehicles. To fill this gap, article, we propose wide-attention deep-composite (WADC) model, consisting module module, article. In particular, can extract global key features flows via linear model with self-attention mechanism. generalize local convolutional neural network component long short-term memory component. We also extensive experiments different datasets to investigate WADC model. Our experimental results exhibit that outperforms existing approaches.