A Data-driven Situation Awareness Method Based on Random Matrix for Future Grids

摘要

Data-driven methodologies are more suitable for a complex grid with readilyaccessible data when tasked with situation awareness. However, it is achallenge to turn the massive data, especially those with some spatial ortemporal errors, into the driving force within tolerable cost of resources suchas time and computation. This paper, based on random matrix theory (RMT),outlines a novel data-driven methodology. 1) Background information andprevious work are reviewed. 2) Related to the methodology, the technical routeand applied framework, data-proceeding and each procedure, evaluation systemand related indicator set, and the advantages over classical methodologies arestudied. Moreover, we make a comparison with the data-driven methodology basedon Principal Component Analysis (PCA). 3) Related functions, including anomalydetection, spectrum test, correlation analysis, fault diagnosis and location,statistical indicator system and its visualization (i.e. 3D power map), aredeveloped. This methodology gains insight into the large-scale interconnectedgrid in a more precise and natural way, it is model free requiring no knowledgeabout the physical model parameters. The methodology, in a flexible andholistic way, processes massive data in the form of large random matrix todepict a global but not a local picture of the system. Meanwhile, the largedata dimension $N$ and the large time span $T$, from the spatial aspect and thetemporal aspect respectively, benefit the engineering performance of theproposed methodology, for this paper, the robustness against unsynchronizeddata is highlighted.