Filling Missing Value Method for Power Quality Data Based on Correlation Analysis

A large number of monitoring indicators own strong correlation among them which help to better fill missing values in these sensor data. In this study, we propose an electric power quality data filling method based on correlation analysis. Firstly, normalized mutual information method is applied to deal with nonlinear correlation which makes up for the deficiency that the traditional Pearson correlation coefficient. Additionally, the measurement of correlation is calculated to obtain the closely correlated indicators. This study utilizes the regression model to build the strong regression model. Experimental results show that the approach can effectively improve the accuracy of filling, reduce the filling error, and improve the quality of data.