Power Systems Dynamic Security Assessment Using Fisher Information Metric

In recent years, complex operating conditions have greatly reduced the predictability of electric grid operations and hence, there is an urgent need to improve grid security more than ever before. The best approach would be to improve grid intelligence rather than simply hardening the grid. An implementation of dynamic security assessment (DSA) would require carrying out time-domain simulations that are computationally too involved to be performed in real-time. This paper presents an approach using machine learning (ML) techniques that would enable the grid to assess its current dynamic state instantaneously. A database consisting of steady-state operating points of the power system and outputs of time-domain simulations is generated in order to train and test the algorithm. A few operating points termed as "landmarks" are identified through a ranking methodology proposed in this paper. Finally, it is shown that prediction accuracy improves through use of such landmarks.