Matrix Filling-based Power Quality Data Restoration System for Power Internet of Things

In the power system under the management of Internet of Things (IoT), some equipment or communication failures may lead to missing and abnormal power quality data, and it is becoming more and more common to repair these missing or abnormal data. Most of the traditional data repair methods are difficult to effectively capture the complex relationship between data, and often have the defect of low repair accuracy. In this regard, this article investigated the power quality data repair system based on the matrix filling method, and utilized a low-rank matrix filling algorithm to deal with large-scale data missing in order to improve the repair accuracy and efficiency. In the study of this article, the system includes three main steps: data preprocessing, matrix filling and result validation. In the data preprocessing stage, the z-score normalization approach is used to carry out the normalization process of the raw data. Singular Value Decomposition (SVD) algorithm is selected to decompose the data matrix into singular value and singular vector matrices with low-rank approximation, which can play a key role in matrix filling. The result validation phase assesses the filling effect through cross-validation and accuracy assessment using error assessment metrics. The experimental results show that at 10% missing rate, the obtained mean square error was about 0.1. At low missing rate, the matrix filling method outperforms the traditional method in several error metrics. The method proposed in this article effectively improves the accuracy and efficiency of power quality data restoration, and provides reliable support for monitoring and control of power IoT.