Statistical method for identification of sources of electromechanical oscillations in power systems

Abstract

The use of real-time continuous dynamics monitoring often indicates dynamic behaviour that was not anticipated by model-based studies. In such cases it can be difficult to locate the sources of problems using conventional tools. This thesis details the possibility of diagnosing the causes of problems related to oscillatory stability using measurement-based data such as active power and mode decay time constant, derived from system models. The aim of this work was to identify dynamics problems independently of an analytical dynamic model, which should prove useful in diagnosing and correcting dynamics problems.

New statistical techniques were applied to both dynamic models and real systems which yielded information about the causes of the long decay time constants observed in these systems. Wavelet transforms in conjunction with General Linear Models (GLMs) were used to improve the statistical prediction of decay time constants derived from the system. Logic regression was introduced as a method of establishing important interactions of loadflow variables that contribute to poor damping.

The methodology was used in a number of case studies including the 0.62Hz Icelandic model mode and a 0.48Hz mode from the real Australian system. The results presented herein confirm the feasibility of this approach to the oscillation source location problem, as combinations of loadflow variables can be identified and used to control mode damping. These ranked combinations could be used by a system operator to provide more comprehensive control of oscillations in comparison to current techniques.