Building low-dimensional damping predictors of the power system modes of oscillation

Abstract

This paper proposes a method for building damping predictors of the power system modes. This method is able to accurately predict the damping of each mode of oscillation and also to locate the variables, monitored by operators, having the biggest influence on the mode dynamics. The method consists of three steps. First, a database of historical data is set up for each potentially dangerous mode of oscillation. This database contains as inputs a set of operating conditions and as output the corresponding mode damping ratio. Afterwards, the inputoutput pairs "operating conditions -mode damping ratio" are used to build an ensemble of regression trees whose structure is analyzed to identify the most relevant variables. In the last step, a second tree-based regressor is run by considering only as inputs these most relevant variables in order to have a lowdimensional predictor. The usefulness of the predictors to avoid poorly damped operating points is then presented. The approach is tested on a 16-machine power system and gives good results.

Inter-area oscillations: Identifying causes of poor damping using phasor measurement units

“…Some authors have already proposed methods to evaluate the power system variables affecting the most the damping. Techniques such as neural networks [7], multiple linear regression analysis [8], generalized linear models [9] and regression trees [10] have shown good results in simulation. However, their application on a large network involves the treatment of a large amount of data.…”