Locating Generators Causing Forced Oscillations Based on System Identification Techniques

Abstract

This paper presents a method to locate the source of forced oscillations in power grids caused by maloperation of turbine-governor systems. The transfer function between the grid frequency and the output active power is first estimated for each generator in normal operation condition. The turbinegovernor acting as the source of the forced oscillations is then detected by identifying rather large mismatches between the predicted output of its estimated transfer function and actual measurements from the grid. The proposed approach can locate both single and concurrent sources of forced oscillations. The method requires synchronized measurements of the grid frequency and the active power of the generator, preferably from PMU measurements. Results from simulations of the Nordic 44 grid model, which emulates the Nordic power system, demonstrate the validity of the proposed method.

Data-based model validation for locating the source of forced oscillations due to power plant governors

“…In case the governor of this particular plant is causing the forced oscillation, the relation between the electrical power and rotational speed of the rotor (a measurement of the electrical frequency is typically used) at this plant will contain a sinusoidal residual, whereas, the residual for the plants withoud forced oscillations will be filtered white noise. This observation is used in the papers [9]- [11]. All these papers require a model of the relation between two measured power system quantities.…”

“…Even though this approach is elegant, the obtained model could suffer from a (relatively) high variance and a bias in the presence of an additional process disturbance. In this paper putting further the intuitive ideas in our previous contribution [9], we use time-domain prediction error identification method that we intorduced in [12] to derive a parametric model of the relation between electrical power and frequency at generator bus bars. This model, which is under the form of a causal transfer function, that is generally more acurate than non-parametric models, allows to compute the residual in the time-domain and to detect the presence of a forced oscillation through the periodogram of the residual.…”