Detection of Bad PMU Data using Machine Learning Techniques

This project presents a pre-State Estimation method for the detection of Bad Data (BD) in Phasor Measurement Units (PMUs) using correlation analysis and a Neural Network Classifier. Presented in this paper is the algorithm design, the steps for generating training and testing data, and the metrics used for evaluation. It is shown that the algorithm is able to detect noisy BD in measurements with load variations, low-level natural noise and transients from bus faults. The benefits of the proposed algorithm include that it can be applied to several measurement subsets in parallel, it is data driven and therefore independent of network model errors, and it is computationally fast, making it a promising technique for online detection.