A Forecasting-Residual Spectrum Analysis Method for Distinguishing Forced and Natural Oscillations

In order to effectively implement remedial reactions to mitigate the negative impacts of oscillations on a power grid, it is essential for system operators to timely and accurately determine whether an observed oscillation is a natural oscillation or a forced oscillation. Using phasor measurement unit (PMU) data, this paper proposes a residual spectral analysis (RSA) method to distinguish forced oscillations and natural oscillations. The proposed RSA method uses forecasting models with various lead times to forecast the current PMU data based on the past PMU data. The spectra of the forecasting residuals are shown to have different properties when the PMU data have forced or natural oscillations. Support vector machines are applied to the residuals to classify the oscillations. This paper develops an algorithm for implementing the RSA method, and demonstrates its superior performance via extensive simulations over the 48-machine model. Simulations show that it can distinguish the oscillations more accurately than an existing method and work reliably even when the frequency of forced oscillations is close to that of natural oscillations.