A Low-Rank Tensor Train Approach for Electric Vehicle Load Data Reconstruction Using Real Industrial Data

Abstract

As electric vehicles (EVs) gain popularity, their interaction with the power system cannot be overlooked. Therefore, there is a growing need for accurate EV load data to facilitate precise operation and control in power systems. However, in practice, due to the high cost of high-frequency measurement devices and limited data storage capacity, only low-resolution metered EV data are available. To address this, this paper proposed a tensor completion-based method for EV load data reconstruction. More specifically, we first reformulate the load data as high-dimensional tensors and consider unknown data to be recovered as missing entries. Subsequently, we leverage the low-rank properties of high-dimensional data to perform tensor completion. To achieve this, two optimization formulations are proposed: a nuclear norm minimization algorithm based on singular value thresholding (SVT) and a tensor rank approximation algorithm via parallel matrix factorization. Both approaches are based on the tensor train (TT) rank, thanks to its well-balanced matricization scheme. This enables us to cost-effectively reconstruct high-resolution EV data using only low-resolution measurements. Simulation results using real industrial data reveal the excellent performance of the proposed methods.