Machine Learning enabled Missing Measurement Data Detection and Recovery of Electricity Grids

Abstract

This paper proposes a machine learning enabled missing data detection and recovery of electrical measurements based on the improved CPCAE. The proposed solution firstly accurately models the missing generation process to generate the missing mask and then combines the absolute difference sequence and the linear correlation as criteria to detect the possible missing segments under different signal-noise ratios (SNR). The solution divides the detected missing mask into different grades and reshapes the origin of one-dimensional data and mask into two-dimensional matrices as a kind of data enhancement. Then we intuitively turn to the deep learning technologies on image processing and design an improved CPCAE model to repair the damaged images. The proposed machine learning-enabled missing data detection and recovery solution are assessed through simulations and the numerical results confirmed its effectiveness for different missing situations.