Bad Data Detection in PMU Measurements using Principal Component Analysis

This paper presents an approach for bad data detection in PMU measurements during disturbances. Bad data in the form of outliers can have very similar appearance as that of system response following disturbances like faults. A principal component analysis (PCA) based approach is proposed to distinguish between the outlier caused by bad data from those caused by disturbances. The principal components (PCs) in the lower dimensional subspace capture the dynamical properties of the system and the PCs in the higher dimensional subspace represent noisy information. Using this property, it is hypothesized that outliers due to bad data will result in larger activity in the high dimensional subspace. Monte Carlo simulation results demonstrate the effectiveness of the proposed hypothesis in an example power system.