Traffic flow tensor prediction framework integrating nonlocal self-similarity features and global consistency

With the vigorous development of intelligent transportation systems, data-driven traffic flow forecasting has played a crucial role in alleviating traffic congestion and reducing traffic accidents. In recent years, the completion algorithm based on tensor has been widely used as an effective traffic prediction method. However, most of the methods only consider the global low rank structure of data and ignore the nonlocal spatio-temporal self-similarity prior characteristics, resulting in low prediction accuracy. This paper proposes a traffic flow tensor prediction framework(NSS-TENSOR) that integrates nonlocal self-similarity features and global consistency of traffic data. A novel traffic flow tensor with strong spatiotemporal correlation is reconstructed by using the dual advantages of nonlocal spatiotemporal self-similarity prior and tensor structure can automatically learn the global consistency of traffic data. Firstly, the nonlocal self-similarity prior characteristics of traffic flow data are captured by clustering algorithm and K-nearest neighbor (KNN) algorithm. Then the pattern information blocks are matched based on the prior features of nonlocal self-similarity, considering the global consistency information simultaneously, a new brief tensor is reconstructed. Through experiments on two real-world datasets, the results show that the proposed method in this paper has higher traffic flow prediction accuracy compared to the original tensor algorithm model.