Robust oscillatory stability assessment for large interconnected power systems

Abstract

This paper deals with robust dynamic security assessment for large interconnected power systems. Special interest is focused on the prediction of critical inter-area oscillatory modes of power systems based on neural networks. After selection of inputs for the neural network and proper training, the stability condition of the power system can be predicted with high accuracy. Hereby, the neural network outputs are assigned to activations of sampling points in the complex plain depending on the distances to the eigenvalues. This method depends highly on the reliability of the measured input data. Missing or bad input data will automatically lead to false prediction results. This paper proposes different methods, which improve the prediction robustness by detecting bad data inputs and outliers. In a second step, input signals identified as bad data inputs will be restored to their correct value