摘要:AbstractThe phenomenon of missing data in traffic has a great impact on the performance of Intelligent Transportation System (ITS). Many imputation methods have been proposed to estimate the missing traffic data. Recently,a tensor-based traffic volume imputation method has been proposed. In this paper, we focus on the underlying mechanism of tensor-based method from the viewpoint of intrinsic multi-correlations/principle components of the traffic data, and try to recommend suitable tensor pattern for traffic volume imputation. Experiments on PeMS database show that the tensor-based method outperforms matrix-based methods, and using the recommended tensor pattern achieves better performances.
关键词:tensor completion;traffic data imputation;principal component analysis;single value decomposition