A study of frequency prediction for power systems

A frequency predictor is identified from simulated tie-line measurements of frequency and power. The performance of the predictor on the simulation and on actual measurements is shown to be identical out to a one second prediction interval. Frequency prediction beyond one second is shown to be poor because the predictor no longer identifies the dynamic relationship between frequency and power and because good frequency prediction at greater than one second ahead of real time would require accurate prediction of random load changes. The predictor is not sensitive to changes in sampling rate and to measurement noise levels as long as the measurement noise is very small with respect to the excursions being measured. Finally, the predictor is tested on different size and types of disturbances. The error in prediction is shown to be proportional to the size of the disturbance. The predictor has large errors for several seconds after the loss of a generator or major tie-line. The predictor tracks the new system dynamics and resumes accurate frequency prediction approximately five seconds after the disturbance occurred.