Distribution System Bad Data Detection Using Graph Signal Processing

As advanced metering infrastructure becomes increasingly prevalent, new data-driven techniques for distribution system control and monitoring have emerged. The efficacy of these techniques, however, can be compromised by erroneous data. This paper develops a graph signal processing-based bad data detection algorithm. First, a physics-based graph construction algorithm is proposed for three-phase power distribution systems. Then, we introduce a data-driven algorithm for detecting erroneous data points by clustering the low-dimensional representations of the graph Fourier transforms of voltage signals. The numerical study results using the IEEE 13-bus test feeder and real-world smart meter time series data show that our proposed algorithm achieves high F1 and accuracy scores.