Nonnegative Low-Rank Tensor Completion Method for Spatiotemporal Traffic Data

Although tensor completion theory performs well with high data missing rates, a lack of attention is encountered at the level of data completion non-negative constraints, and a remaining lack of effective non-negative tensor completion methods is still found. In this article, a new non-negative tensor completion model, based on the low-rank tensor completion theory, called the Nonnegative Weighted Low-Rank Tensor Completion (NWLRTC) method, is proposed. Due to the advantages of Truncated Nuclear Norm (TNN) in low-rank approximation, NWLRTC considers the TNN as the objective optimization function and adds a directional weight factor to the model to avoid its dependency on the data input direction. In addition to considering the completion accuracy, NWLRTC also imposes non-negativity constraints to meet the requirements of practical engineering applications. Finally, NWLRTC is realized by the alternating direction multiplier method. As for the experiments, they are carried out using different methods for generating missing data and for different iteration times. The experimental results show that the NWLRTC algorithm has high completion accuracy at low missing data rates, and it maintains a stable completion accuracy even when the missing rate hits 80%.