Enhanced PMU data imputation using variational autoencoder: a novel approach for improved accuracy

Data collected from various measurement devices, such as phasor measurement units (PMUs), form the backbone of power system monitoring algorithms such as state estimation algorithms. Even though modern PMUs are efficient, data inaccuracies and loss can still occur due to problems like equipment malfunctions, communication issues, and hardware or software glitches. Missing values in the datasets can also affect the overall quality, impacting the algorithms dependent on them. This article introduces a novel technique for real-time imputation of missing data using variational autoencoder (VAE) and utilizing the latent space representation acquired from VAE. The method has demonstrated enhanced accuracy in imputing missing data, surpassing traditional and existing techniques. Its efficiency was validated by testing the algorithm on real-world PMU data from Power System Operation Corporation Ltd., India (recently renamed Grid Controller of India Limited), showing significant improvements in imputation accuracy. These f indings illustrate the potential of using VAE techniques to address missing PMU data in power system monitoring, leading to improved data analysis and decision-making. The proposed algorithm has shown exceptional results in both root mean square error and mean absolute error metrics for various percentage of missing PMU data.