Probabilistic Deep Autoencoder for Power System Measurement Outlier Detection and Reconstruction

Abstract

A probabilistic deep autoencoder is proposed to reconstruct power system measurements in this paper, which can be utilized in outlier detection and reconstruction. A nonparametric distribution estimation method is employed to capture the uncertainty information of the measured data. The estimated confidence intervals of the measured data are extracted from the estimated distribution and used as input to the first layer of neural networks. Through multilayer encoding and decoding processes, the intervals of measurements are reconstructed, which are further applied to detect and replace outliers. Simulation results verify the effectiveness of the proposed method.