Missing data imputation considering multi-mode variations

Missing traffic data are inevitable due to detector or communication malfunctions which adversely affect the performance of intelligent transportation systems and make the requirement of missing traffic data imputation more important. In this paper, a novel method based on tensor completion is proposed to estimate the missing traffic data. Compared with previous tensor-based methods, systematic variations encoded with total variation are used to mine the traffic intrinsic properties. By minimizing the total variation norm, the approach can keep the systematic variations of traffic volume while inheriting the advantage of mining the multi-dimensional correlations of traffic data from the tensor pattern. Experimental results on PeMS database show the proposed method achieves a better imputation performance than the state-of-the-art missing traffic data imputation approaches.