Data compression for advanced monitoring infrastructure information in power systems based on tensor decomposition

• Published in: Sustainable Energy, Grids and Networks Grids and Networks, 2022-12, Vol.32, p.100917, Article 100917
• Main Authors: Sandoval Guzmán, Betsy, Barocio, Emilio, Korba, Petr, Obushevs, Artjoms, Segundo Sevilla, Felix Rafael
• Format: Article
• Language: English
• Subjects: Clustering; Data compression; Missing data; PMUs; Smart meters; Tensor decomposition; Tucker decomposition
• ISSN: 2352-4677; 2352-4677
• DOI: 10.1016/j.segan.2022.100917

In this paper, an innovative data compression methodology based on Tucker tensor decomposition is presented. The proposed approach exploits the benefits of organizing time series in multidimensional arrays (tensors) in order to achieve higher compression ratios of the original data while preserving most of their properties and, in this form, achieving a low reconstruction error. This reorganization of the data into the tensor allows the creation of correlation for heterogeneous data. Furthermore, it reduces the direct relationship between the number of columns and rows with the final size of the compressed file when the data is compressed using algorithms based on matrix-based dimensional reduction techniques, such as the Singular Value Decomposition (SVD). At the same time, the problem of missing data can be overcome since the problem is formulated as an iterative process. To demonstrate the effectiveness of the innovative algorithm proposed, two study cases from electrical networks are presented. First, the compression of real synchrophasor data acquired from one commercial Phasor Measurement Unit (PMU) measuring 3-phase voltages and frequency measurements from the electrical system of Continental Europe is presented, where a significant CR is achieved for apparently no correlated data. Then, an application to real Smart Meter measurements from the electrical utility ERCOT in the USA is also provided. Here the clustering of the data is obtained as an additional outcome of the proposed algorithm. To underline the benefits of the innovative proposed methodology in contrast to other more common decompositions such as SVD, the compression ratios and reconstruction error achieved using these two different techniques are compared. The results indicate that using Tucker decomposition, not only higher compression can be achieved, but also it is demonstrated that for SM data, additional attributes can be obtained, such as clustering.