Power Frequency Estimation based on a Bayesian Online Changepoint Detector

When the voltage and current signals measured in ac power systems are highly distorted due to high penetration of nonlinear loads and power electronics converters (e.g., in microgrids) and measurement noise, the traditional methods of frequency calculation will be associated with a significant error. This article introduces a novel power frequency estimation method based on a Bayesian online changepoint detector. In each time step of this method, the most efficient changepoints are detected in a window containing the latest measured voltage samples. The frequency is estimated and updated after applying some refining procedures on these changepoints. The results of numerical studies on various simulated and laboratory-generated signals under standard and severe conditions confirm that the suggested method can estimate the frequency accurately and robustly in dealing with frequency deviations, harmonics, interharmonics, dc components, noise, nonideal sampling, amplitude and phase angle modulations, and step variations. For instance, the proposed method can tolerate high noise levels and remain robust even at a signal-to-noise ratio of 20 dB, achieving mean absolute errors of 0.036 Hz without prefiltering and 0.004 Hz with prefiltering. It can also exhibit an invariant response to significant step variations in the voltage amplitude and phase angle.