计算机科学
联营
空间分析
人工智能
同步
变压器
数据挖掘
电压
遥感
电信
物理
量子力学
传输(电信)
地质学
作者
Chu Wang,Jiadi Hu,Ran Tian,Xin Gao,Zhongyu Ma
标识
DOI:10.1007/978-3-031-30637-2_27
摘要
As a typical problem in spatial-temporal data learning, traffic prediction is one of the most important application fields of machine learning. The task is challenging due to (1) Difficulty in synchronizing modeling long-short term temporal dependence in heterogeneous time series. (2) Only spatial connections are considered and a mass of semantic connections are ignored. (3) Using independent components to capture local and global relationships in temporal and spatial dimensions, resulting in information redundancy. To this end, we propose Inception Spatial Temporal Transformer (ISTNet). First, we design an Inception Temporal Module (ITM) to explicitly graft the advantages of convolution and max-pooling for capturing the local information and attention for capturing global information to Transformer. Second, we consider both spatially local and global semantic information through the Inception Spatial Module (ISM), and handling spatial dependence at different granular levels. Finally, the ITM and ISM brings greater efficiency through a channel splitting mechanism to separate the different components as the local or a global mixer. We evaluate ISTNet on multiple real-world traffic datasets and observe that our proposed method significantly outperforms the state-of-the-art method.
科研通智能强力驱动
Strongly Powered by AbleSci AI